首次运行时即可启动的开源 MCP 服务器

Playwright MCP 服务器

测试发现：The Playwright MCP server v0.0.80 installs, builds, passes all 8 tests (browser navigation, click, capabilities, CLI, and CommonJS import), and responds to --version and --help correctly using Chromium in headless mode. 6 分钟。

社区插件，可使用你选择的任何 LLM 来控制 Blender 3D。与官方 Blender Foundation 无关。

测试发现：mcp-for-blender v2.0.0 builds with uv, all 152 tests pass, and the MCP server binary starts and correctly responds to JSON-RPC initialize requests; the install-addon CLI reports Blender not found (expected, Blender is not installed in this container), and no third-party API keys were needed for any test to pass. 2 分钟。

面向机器人和 agent 的 AI 搜索引擎与 MCP 服务器。单个 Rust 二进制文件即可索引代码、文档、日志、JSONL/CSV 和 PDF，用于 BM25 全文、向量和混合搜索、无需 grep 的代码搜索、agent 记忆与 RAG。兼容 Elasticsearch 的 API；Corpus Hub...

测试发现：XERJ v1.0.0-rc.77 built from source in 2m06s (quick profile), serving ES-compat API on :9200 with autoindex and MCP server working; test suites pass except 2 IPv6 bind failures specific to this container. 36 分钟。

将文档网站、GitHub 仓库和 PDF 转换为 Claude AI 技能，并自动检测冲突

测试发现：skill-seekers 3.10.0.dev0 installed and all 3857 unit/adaptor/scraper tests pass without external API keys or network dependencies. 6 分钟。

无限制的浏览器自动化与测试。SeleniumBase 的 CDP Mode 可轻松绕过机器人检测。查看示例！

测试发现：SeleniumBase 4.54.7 is installed in a Python 3.12 venv; 8 framework unit tests, 7 offline HTML tests, and 2 real web tests passed using headless Chrome 154. 5 分钟。

将你的 FastAPI 端点作为 Model Context Protocol (MCP) 工具暴露，并支持 Auth！

测试发现：All 89 tests pass, MCP SSE transport, HTTP transport, OAuth config, and OpenAPI conversion are exercised with real MCP client/server communication over the wire using a live FastAPI app. 10 分钟。

为团队中的每个 coding agent 提供一份共享记忆。Claude Code、Cursor 和 Codex 会获得你代码和团队过往决策的地图，从而从你停下的地方继续。

测试发现：The graft CLI installs, builds, and its full test suite passes (1491/1492, 1 deliberate skip) with both previously-failing shim-resolution and lock-release tests green, and the CLI builds/parses/queries a scratch repo and completes an LLM deep pass against a local mock endpoint. 87 分钟。

团队中的每个 coding agent 共享一个记忆。Claude Code、Cursor 和 Codex 获得你的代码和团队过往决策的地图，从而从你停下的地方继续。

测试发现：npm install completes, TypeScript compiles, the CLI builds/checks/queries a structural code graph on a real repo without any keys, and all 1220 tests run with 0 failures. 10 分钟。

终结代码幻觉。GitMCP 是一个免费的、开源的远程 MCP 服务器，适用于任何 GitHub 项目。

测试发现：The GitMCP app installs, builds, and serves its web UI on the react-router dev server at HTTP 200, and all 37 unit tests pass. 7 分钟。

用于 AI coding agents 的持久化记忆系统。与 agent 无关的 Go 二进制程序，支持 SQLite + FTS5、MCP 服务器、HTTP API、CLI 和 TUI。

测试发现：Engram v2.0.0 builds from source, all 29 test packages pass, CLI commands (save/search/context/timeline/doctor/test) work, and the HTTP API (health/sessions/observations/search) returns correct responses with real SQLite storage. 8 分钟。

直接从 Cursor 和其他兼容 MCP 的 IDE 监控浏览器 logs。

测试发现：The BrowserTools MCP workspace installs, builds, and passes all 356 unit and integration tests with Node 22.14.0. The MCP server binary prints version 2.0.2, help with 16 tools, and runs --doctor setup checks. 12 分钟。

Browser MCP 是一个 Model Context Provider (MCP) 服务器，允许 AI 应用控制你的浏览器。

测试发现：All dependencies install, the project builds, typechecks, and all 252 tests pass across all 4 tested packages using real headless Chromium, and the MCP server binary responds to --help. 8 分钟。

一个 Model Context Protocol (MCP) 服务器和 CLI，用于在开发 iOS 和 macOS 项目时为 agent 提供工具。

测试发现：MobileBuildMCP 2.7.1 builds, typechecks, and passes all 2591 unit tests (posttest also passes: 16/16 warden-watchdog tests). The CLI binary responds with version 2.7.1, lists 72 canonical tools, and starts the MCP server on stdio. All of this was done on Linux (no Xcode/macOS), so the macOS/Xcode-specific tools cannot... 10 分钟。

用于与 Godot 游戏引擎对接的 MCP 服务器。提供用于启动编辑器、运行项目和捕获调试输出的工具。

测试发现：Install builds cleanly. MCP server starts and responds to all 13 tools. Real Godot v4.3-stable binary was downloaded and used for verification, 10 tools work end-to-end, 2 (load_sprite) fail due to headless rendering limitation, and 2 (get_uid, update_project_uids) correctly report needing Godot 4.4+. 11 分钟。

对 Gemini Notebook 的编程访问：通过命令行界面 (CLI)、Model Context Protocol (MCP) 服务器和 AI agent 技能。

测试发现：Dependencies installed, both CLIs (nlm, notebooklm-mcp) produce help output, MCP server binds to HTTP port, and 1899/1899 unit tests pass. 8 分钟。

面向 agent 的快速准确代码搜索。相比 grep+read 减少 99% 的 token 使用量。

测试发现：semble 0.5.5 installed in a virtual environment at /work/repo/.venv; all 330 tests pass; CLI and MCP server run correctly on CPU. 2 分钟。

你的浏览器就是 API。为 AI agent 提供 CLI + MCP server，以使用你的登录状态控制 Chrome。

测试发现：bb-browser pnpm install and build succeed; CLI version 0.14.2 prints help; daemon auto-downloads and launches Chrome 149 headed on Xvfb; HTTP API on port 19826 responds 200 to /status, tab_list, eval, snap, open commands; 129 of 143 tests pass, 13 skipped (need Chrome), 1 flaky timing failure. 22 分钟。

面向 AI 助手的市场分析 MCP 服务器：加密货币、股票、外汇和期货的筛选器、多周期技术分析、回测和情绪分析。提供 39 个工具。可免费自托管，或使用托管版本 TradingMCP。

测试发现：The tradingview-mcp-server Python package is installed in a .venv at /work/repo, all 286 unit tests pass, and the MCP server starts and responds to initialize, tools/list (37+ tools), and tools/call on stdio transport. 2 分钟。

官方 Notion MCP 服务器

测试发现：npm install completes, npm run build produces bin/cli.mjs, all 109 tests pass, and the HTTP server starts and responds to health checks and JSON-RPC requests on the MCP endpoint. 3 分钟。

用于 LinkedIn 、 Indeed 、 Glassdoor 、 ZipRecruiter 、 Bayt 、 Naukri 、 BDJobs 和 HelloWork 的职位抓取库

测试发现：JobSpy 1.4.0 installs, builds, imports, and runs: the CLI searches Indeed and Glassdoor for real jobs, the library scrapes all 9 job boards, the MCP server responds over stdio JSON-RPC, lint and type-checking pass cleanly, and both pip and poetry workflows succeed. 4 分钟。

Flint 是一种可视化语言，它让 AI agent 能够从简单且可人工编辑的图表规格中可靠地创建富有表现力且美观的图表。

测试发现：npm install succeeds, all 1739 Vitest tests pass, both JS library and MCP server build to working dist output with functioning CLI binaries. 14 分钟。

面向 coding agents 的本地视频编辑器：42 个 FFmpeg 工具，带有机器可读的合约和一个 MCP 服务器，适用于 Claude Code、Cursor 和 Codex。剪辑、字幕、重新构图、音量、平台导出。无需云服务，无需 API 密钥。

测试发现：ffmpeg-skill 2.10.3 installed cleanly into /tmp/skilltest/ffmpeg-skill, doctor reports 67 capabilities with 0 required missing, the 169-test contract suite exits OK, and a synthetic clip was cut, probed at 75 frames/2.52s and decoded with no ffmpeg errors. 25 分钟。

特定于产品的 Cloudflare MCP 服务器

测试发现：Monorepo installs cleanly with Node 22; full test suite passes; demo-day MCP server starts locally and answers Streamable HTTP tools/list requests. 6 分钟。

一个简单、安全的 MCP 到 OpenAPI 代理服务器

测试发现：mcpo installs and runs correctly, proxying a real stdio MCP server (FastMCP) over REST with auto-generated OpenAPI schemas; all 27 unit tests pass, both tool endpoints return correct results via HTTP. 5 分钟。

一个可视化 MCP 与技能，包含 25+ 个使用 @antvis 的可视化图表，用于图表生成和数据分析。

测试发现：All 43 tests pass, the MCP server builds and starts on stdio/SSE/streamable transports, and generates real chart images via the AntV GPT-Vis API. 38 分钟。

一个 MCP 服务器以及一个 CLI 工具，用于将本地代码索引到图数据库中，为 AI 助手提供上下文。

测试发现：CodeGraphContext 0.6.13 installed and working: CLI, FalkorDB/KuzuDB/LadybugDB backends, indexing, MCP server, and 1615 of 1627 tests pass. 36 分钟。

Excel MCP server：让 AI（Claude、Cursor、Codex）创建和编辑包含图表、数据透视表、切片器和公式的 Excel 文件。无需 Excel。

测试发现：Install builds cleanly via pip install -e . in a venv. All 5 existing tests pass. The MCP server starts in stdio, SSE, and streamable-http modes. Tools/list returns all 25 tools, and create_workbook/write_data/read_data work end-to-end via the MCP stdio protocol. 4 分钟。

全能协议分析工具：内置浏览器抓包、MITM 代理、JS hooks、指纹伪装、AI 分析及 MCP server，用于 agent 集成。

测试发现：Installation, build, full test suite (185/189 passing), Electron binary, and built app all work in the container. Only expected dbus/GPU errors appear in container. 9 分钟。

用于 Model Context Protocol 服务器和客户端的官方 Java SDK。与 Spring AI 合作维护。

测试发现：MCP Java SDK compiles and all 771 non-Docker tests pass with 0 failures; 12 Docker-dependent tests are skipped. 15 分钟。

专注于 Laravel 的 MCP 服务器，用于增强你的 AI 驱动的本地开发体验。

测试发现：Laravel Boost library installed with PHP 8.3.33 and Composer 2.10.3; all 1100 tests pass (3257 assertions) and all 5 architecture checks pass. 8 分钟。

面向 AI agents 的隐身浏览器自动化，基于 MCP 和 Chrome DevTools Protocol：导航、网络钩子、DOM 提取和像素级准确的 UI 克隆。

测试发现：Stealth Browser MCP server runs on stdio transport, spawns a headless Chromium 157.0.8086.0, navigates to data URLs, extracts page HTML content, and closes cleanly -- all via MCP protocol over stdio. 15 分钟。

Cortex：从 OpenAPI、AsyncAPI、GraphQL、gRPC、OpenRPC 和 Markdown 生成交互式 API 文档、类型化 SDK 和 MCP 服务器。

测试发现：Cortex Docs installs, builds, passes its full unit test suite (377 tests, 0 failures), and the CLI successfully initializes a project, validates specs, and generates SDKs in 11 languages plus an MCP server from OpenAPI, AsyncAPI, GraphQL, and OpenRPC sources. 6 分钟。

Neo.mjs 是一个自我进化的软件有机体：一个专业的端到端 AI 工程团队，其跨模型集群通过 Neural Link、Active Hybrid GraphRAG、DreamService 和自愈循环驻留在实时应用中。

测试发现：Project installed, bundled browser deps, built CSS themes, and ran all 4662 unit tests passing with no failures. 6 分钟。

一个用于构建 MCP 服务器的 TypeScript 框架。

测试发现：The project builds and 685+ tests pass across auth, edge, bin, openapi, and FastMCP sub-suites. The edge WebStreamableHTTPServerTransport now correctly imports `randomUUID` from the `crypto` module instead of relying on the global `crypto` variable. Remaining failures are pre-existing: openapi loadSpec hits Node 18... 57 分钟。

开源 SMS 网关。将你的 Android 手机变成 SMS API：通过 REST 和 webhooks 从你的应用发送和接收 SMS。可自托管，并提供免费托管方案。

测试发现：The textbee.dev monorepo builds and runs: API (NestJS) compiles and initializes all modules; Web (Next.js) builds all routes; all 67 API test suites (1478 tests) and all 49 web test files (428 tests) pass. 24 分钟。

一个用于 agent 文献工作的本地 MCP 服务器。支持原始 LaTeX 章节阅读、从 arXiv 元数据生成 BibTeX，以及主题关注。论文保留在磁盘上。搜索为可选功能。

测试发现：The arxiv-mcp-server Python package builds, all 447 tests pass, and the server launches via HTTP transport responding 200 on /healthz without any modifications to the repository. 2 分钟。

一个 MCP 服务器，用于让 LLMS 获取关于 shadcn ui 组件结构、用法和安装的上下文，兼容 react、svelte 5、vue 和 React Native

测试发现：The shadcn-ui-mcp-server builds from source, starts in both stdio and SSE modes, responds to MCP protocol requests, and fetches real component data from the shadcn-ui/ui GitHub repository without any modifications or mocks. 5 分钟。

一个用于将几乎任何内容转换为 Markdown 的 Model Context Protocol 服务器

测试发现：Install (bun install, bun run build) complete. All 79 tests pass. MCP server launches via bun start/bun dist/index.js and responds to tools/list and tools/call over stdio. 9 分钟。

面向 Google Meet、Microsoft Teams 和 Zoom 的开源会议转录 API。自动加入机器人、实时 WebSocket 转录、面向 AI agent 的 MCP 服务器。支持自托管或使用托管 SaaS。

测试发现：11 Python service test suites and the JS terminal client test suite run successfully in a container with Node 22.13 and Python 3.12.3; the compose/Docker deployment path cannot execute without Docker engine being available. 17 分钟。

读取、修改和重写韩国文档的 AI 文档工具。将 HWP、HWPX、PDF、Office、扫描件（OCR）转换为 Markdown，支持新旧对照、保留格式编辑、Markdown 转 HWPX、HWP 转...

测试发现：kordoc 4.13.1 builds, all 1691 automated tests pass, the CLI parses HWPX and XLS files to Markdown correctly, and the library API returns structured parse results with fileType, markdown, blocks, and metadata. 28 分钟。

AI 查找法令和判例并验证引用的 MCP 服务器：将法制处 42 个 API 整合为 10 个工具，覆盖法令、条文、附表，判例、宪法裁判、解释例等 23 个领域，提供条例整备雷达和新旧条文对照表 | 面向韩国法律的 MCP 服务器：支持法令、判例、条例、对照表及引用验证

测试发现：Korean Law MCP server v4.14.2 builds, all 989 tests pass, STDIO and HTTP modes run correctly with MCP protocol, CLI reports version, 10 tools registered via tools/list. 3 分钟。

Streamable HTTP 与 stdio MCP 传输之间的桥梁

测试发现：mcp-proxy 0.12.0 installs cleanly, CLI responds to --version and --help, 91/92 unit tests pass covering CLI argument parsing, config loading, MCP server, HTTP/SSE transports, progress forwarding, and proxy server request forwarding. 3 分钟。

Java 企业级应用开发框架，适用于全场景：克制、高效、开放、生态。并发能力提升 700%，内存节省 50%，启动速度快 10 倍，打包体积缩小 90%；兼容 java8 ~ java26；支持 LTS。（可替代 spring）

测试发现：Solon Java framework: all 23 bundles compile and install; 46 selected unit tests pass cleanly on JDK 17 with Maven 3.9.6. 27 分钟。

集 MCP 聚合器、编排器、中间件、网关于一体的 Docker 镜像

测试发现：MetaMCP backend server starts on port 12009 with PostgreSQL, serves SSE/MCP/OpenAPI endpoints, authenticates users via email/password, bootstraps users/keys/namespaces/endpoints from env, and responds to all standard MCP protocol methods (initialize, tools/list, prompts/list) successfully. 11 分钟。

用于构建 Go 应用的 CLI 工具。

测试发现：Go 1.27.1 installed at /tmp/go, nunu CLI installed at $HOME/go/bin/nunu, source repo builds and all 4 test packages pass, created testapp project compiles and serves HTTP 200 on :8000, hot-reload and wire injection both functional. 5 分钟。

用于 Excalidraw 的 MCP 服务器和 Claude Code 技能：一个可编程的画布工具包，可通过 AI agent 创建、编辑和导出图表，并支持实时画布同步。

测试发现：Backend TypeScript compiles, npm test suite (MCP wire protocol + canvas bind checks) passes 7/7 checks, canvas server runs serving REST API on :3000, CLI commands (add, describe, export, status) all function correctly creating/managing Excalidraw elements. 27 分钟。

生产就绪的 MCP 服务器，支持实时搜索、提取、映射与爬取。

测试发现：npm install succeeds, TypeScript compiles without errors, all 6 unit tests pass, and the MCP server starts and advertises all 6 tools (tavily_search, tavily_extract, tavily_crawl, tavily_map, tavily_research, tavily_feedback) via JSON-RPC over stdio. 2 分钟。

这是一个基于 Model Context Protocol（MCP）的 12306 车票搜索服务器。

测试发现：12306-mcp v0.3.10 builds and runs: stdio MCP server responds with 8 tools, loads live 12306 station data, and the HTTP/SSE server starts on port 8080. 3 分钟。

从任何 AI agent 获取生产就绪的图像、视频、音频和 3D 内容：技能包括选择正确的模型、在花费前定价，并通过 Scenario MCP 保持角色和品牌一致，以及驱动 Blender、Maya、ZBrush、Unreal 和 Unity 的专家团队。

测试发现：The repository installs cleanly with pnpm 9.15.9, all 122 Agent Skills pass Agent Skills spec validation, all shipped-script test suites (Python and TypeScript) run to completion with 0 failures, and the full CI validation pipeline (prettier, cspell, groups, manifest, README sync, spec validation) passes after... 9 分钟。

Skybridge 是一个用于 MCP 应用和 ChatGPT 应用的全栈 TypeScript 框架，类型安全，基于 React，平台无关。

测试发现：Skybridge monorepo installs, builds, and all 476 unit tests pass across all packages (core, vite-plugin, devtools, create-skybridge, test). The CLI reports skybridge/1.0.0 and responds to --version and --help commands. 44 分钟。

面向 AI agent 流水线（LangGraph、CrewAI、AutoGen）和 Claude 的开源持久化内存。REST API + 知识图谱 + 自主整合。

测试发现：mcp-memory-service 11.13.0 installed from source via pip install -e '.[dev]', the uvicorn server starts on 127.0.0.1:9791 (HTTP) with sqlite_vec backend, health endpoint returns 200, memory CRUD operations work end to end, and the full test suite passes except for one benchmark test that fails due to hash-embedding... 20 分钟。

面向大规模工程的语义智能。Context+ 是一个 MCP 服务器，专为要求 99% 准确率的开发者设计。通过结合 RAG、Tree-sitter AST、谱聚类和 Obsidian 式链接，Context+ 将庞大的代码库转化为可搜索的分层功能图。

测试发现：npm install + npm run build succeed; all 271 tests pass; the MCP server starts on stdio and shuts down cleanly. 1 分钟。

面向 AI agent 和 IDE 的高性能代码智能引擎，支持 257 种语言、多仓库、基于图，可通过 CLI、MCP Server 和 API 访问。作为 AI coding agent 的队友，只暴露必要信息，将 token 使用量最多减少 50 倍。100% 本地。

测试发现：Go 1.27.0 installed locally. Gortex builds from source (v0.64.5) and the daemon indexes this repo into a knowledge graph of 182,637 nodes and 822,458 edges. Core graph, search, parser, resolver, query, analysis, semantic, serverstack, and agent test suites pass. The pre-built release binary v0.64.7 was also verified. 69 分钟。

一个用于构建生产级 agent 系统的 Go 框架，支持图工作流、工具、记忆、A2A、AG-UI、MCP、评估和可观测性。

测试发现：The tRPC-Agent-Go root module (trpc.group/trpc-go/trpc-agent-go) builds and all 235 test suites pass, the test/ module passes with Go 1.24.4 auto-downloaded, and the examples/ module builds successfully, all from a fresh container with Go installed from the upstream tarball. 14 分钟。

Co-op Anything：让你的 AI agent 针对任何单人游戏，并与朋友一起玩（在线和/或分屏）。MCP server 和 playbook 让你逆向工程、修复、测试、发布和分享它学到的东西。

测试发现：After creating a project venv (PyYAML) and patching two real defects, the buildlock Windows-only mutation twin not used off-Windows and the MCP detect_engine silently dropping family matches without PyYAML, every Linux-runnable tool suite passes: 12 of 15 test suites ALL OK with gamepool and hotreload skipping by... 31 分钟。

Grounded Docs MCP Server：Context7、Nia 和 Ref.Tools 的开源替代方案

测试发现：Node 22.23.3, npm install, build, and all 2325 tests pass; CLI --help prints correctly. 8 分钟。

用于 70 个巴西公共 API 的 MCP Server

测试发现：All 2566 tests pass, the server responds correctly to MCP initialize (2024-11-05 protocol) with 363 tools registered across 51 features, no code modifications required. 3 分钟。

用于被动 OSINT 检查的开源引擎：21 个工具，以 CLI、MCP 服务器和 Web UI 形式提供。托管 Monitor & Diligence 位于 app.openosint.tech。

测试发现：OpenOSINT 2.29.0 installs from source, CLI shows help with 20 tools, DNS tool resolves real domains, playbook generates real investigation reports, web server serves on port 9877 with HTTP 200 on / and /api/health, and 712 automated tests pass. 7 分钟。

发现灵感，并使用最新的 AI 模型创建图像和视频。可与 ChatGPT、Claude、Cursor 和其他 MCP 客户端配合使用。

测试发现：Project builds and all 128 library + tool tests pass on Node 18.19.1. The MCP server module loads and the CLI init command responds. The release CI script requires Node 22+ and pnpm which are not available in this container. 56 分钟。

Karpathy's LLM Wiki 的开源实现。上传文档，通过 MCP 连接你的 Claude 账户，然后让它编写你的 wiki。

测试发现：Python API serves health endpoint on localhost:8000, CLI initializes workspaces, all unit and integration tests pass on SQLite and Postgres backends. 18 分钟。

可靠的 AI 技能 + MCP 工具包，用于 agent 驱动的桌面 CAD 自动化。

测试发现：The project installs, all 739 tests pass, the MCP server answers initialize/tools/list, open-format export produces valid watertight STL, OBJ, DXF and readable PDF, and the desktop app launches when set to QT_QPA_PLATFORM=offscreen (normal xcb display requires libxcb-cursor0 which needs root). 23 分钟。

一个轻量级的 Model Context Protocol (MCP) 服务器，用于安全地访问 Obsidian vault

测试发现：All 279 tests pass, the server builds and runs correctly, and read_note returns content from a real vault note via MCP stdio transport. 4 分钟。

将仅支持本地（stdio）服务器的 MCP 客户端连接到远程 MCP 服务器。

测试发现：All 490 unit/integration tests pass, all 8 concurrent-instance OAuth tests pass with the local OAuth simulator, all 2 real E2E tests connect to Cloudflare and Hugging Face MCP servers, all 4 protocol-era bridge tests pass, and the dist build runs without toWellFormed errors on Node 18. 40 分钟。

即插即用的库，使 agent 能够通过代码执行调用 MCP 和 UTCP 工具。

测试发现：TypeScript library @utcp/code-mode builds and passes all 29 tests using [email protected] native sandbox. Python library builds and passes all 17 tests using RestrictedPython sandbox. CLI and MCP bridge both start and respond correctly. 17 分钟。

Pi coding agent 的 Token-efficient MCP adapter

测试发现：Installed and running on Node 22.23.3. Tests pass 1997/2000, typecheck passes, public exports verified, CLI works. 9 分钟。

RuleGo 是一个轻量级、高性能、嵌入式、下一代组件编排规则引擎框架，适用于 Go。

测试发现：The rulego Go library compiles with go build and all its test suite passes (go test exits 0, PASS). 8 分钟。

一个命令行界面，用于通过 stdio 和 HTTP 传输与 MCP (Model Context Protocol) 服务器交互。

测试发现：The mcp binary builds and runs, all tests pass, and the mock server, tools listing, and tool calling work end-to-end from the CLI. 13 分钟。

官方的 data.gouv.fr MCP 服务器，允许 AI 聊天机器人通过对话直接搜索、探索和分析来自法国国家开放数据平台的数据集。

测试发现：Python 3.12 environment with relaxed requires-python; package installs and builds; all 147 tests pass; MCP server starts and responds HTTP 200 on /health. 5 分钟。

管理并连接所有 MCP 服务器的统一位置

测试发现：MCPJungle builds and runs in this container: compiled binary responds to CLI commands, server starts on port 8080, health endpoint returns OK, MCP JSON-RPC endpoint responds to initialization, dashboard serves its React UI, and the full test suite passes. 14 分钟。

一个官方的 Qdrant Model Context Protocol (MCP) 服务器实现

测试发现：The MCP server builds, all 24 tests pass, and the server responds correctly over stdio transport with initialize, tools/list, tools/call (store and retrieve using Qdrant :memory: mode with fastembed). 3 分钟。

用于 AKShare 的 HTTP API 和 MCP 服务器：从任何编程语言或 AI 助手（Claude、Cursor）获取金融数据

测试发现：AKTools 0.1.1 installed via pip in a Python 3.12 venv, HTTP server on port 8888 responds to /, /api/interfaces, and /version endpoints with 200 OK. 1 分钟。

麦当劳省钱点餐助手（麦门精算师）｜麦当劳优惠券组合优化技能：拉取麦当劳菜单与全部可用优惠券，算出最省的点餐组合，巨无霸套餐 ¥47 → ¥21.9（省 53%）。McDonalds MCP（麦当劳 MCP）优惠券优化器 · McDonalds China MCP Server · 用于 WorkBuddy 的...

测试发现：All 4 Python scripts (optimizer, selftest, demo, bench) and the Node.js parity check run successfully. 46 self-tests pass. Demo shows 7 scenarios working end-to-end with sample data. 1 分钟。

一个实现与 MySQL 数据库安全交互的 Model Context Protocol (MCP) 服务器

测试发现：MySQL MCP server installs, builds, and runs with all 26 tests passing against a real MySQL 8.0 instance. STDIO and SSE transport modes verified. All MCP tools (execute_sql, get_schema_info, get_table_sample), prompts (explore_database, analyze_table), and resources are functional. 7 分钟。

用于从 deepwiki.com 获取最新知识并在 Cursor 及其他代码编辑器中使用的 MCP server

测试发现：All 16 tests pass, the MCP server builds and runs on both stdio and HTTP transports, and responds to tools/list with the deepwiki_fetch tool. 4 分钟。

用于 Apple 开发者文档的 MCP 服务器：在 Claude、Cursor 和人工智能助手中搜索 iOS/macOS/SwiftUI/UIKit 文档、WWDC 视频、Swift/Objective-C API 和代码示例

测试发现：npm install and tsc build succeed, all 462 tests pass, the MCP server binary launches on stdio and correctly responds to initialize, tools/list, and tools/call with bundled WWDC data (1266 videos, 12 years). 4 分钟。

免费的双向 Figma MCP 服务器。将设计转化为框架感知的代码，并将代码推回画布。适用于 Claude Code、Cursor、Codex 以及任何 MCP 客户端。

测试发现：The repository installs and builds cleanly under Node 24.17.0/pnpm 12.8.0, its full 2162-test suite passes, and the built MCP server launches, advertises 116 tools, and answered a real tools/call ping over stdio. 6 分钟。

安全的 MCP 运行时，让 AI agent 操作本地机器、服务器和容器，支持多设备编排。

测试发现：AgentDock binary was built successfully, all 53 test packages passed (0 failures), the server started on 127.0.0.1:8765, and a JSON-RPC ping to the MCP endpoint returned HTTP 200 with a valid result. 2 分钟。

人类和 AI Agent 的临时房间：无需永久工作空间即可运行、监督和引导 Agent 任务。

测试发现：The web app installs, builds, passes all 1406 unit tests, type-checks with zero errors, and the dev server starts serving HTTP 200 on / and /room. 5 分钟。

[已被 The Guild 收购] The Grafbase GraphQL Federation Gateway

测试发现：Grafbase gateway (v0.53.5) and CLI (v0.105.2) compiled from source; gateway serves GraphQL endpoint on localhost and passes 1364 of 1384 integration tests, with all 19 failures from missing external Docker containers, mTLS cert snapshots, and insta loop assertion limits. 64 分钟。

开源 MCP 服务器，用于比较产品并在购买前询问买家。

测试发现：On Node 22.14.0, install and release:build complete cleanly, the full test suite (pnpm test) exits 0 with 1197 vitest tests passing, and the local-mode MCP server starts on loopback and serves its tools, reporting version 0.2.1. 19 分钟。

80 多个生产就绪的 AI agent 示例（Python）：使用 uAgents、ASI:One、MCP、A2A、LangChain、CrewAI、Gemini、Claude 和 OpenAI 构建自主 agent、多 agent 系统和 agentic AI。

测试发现：The Fetch hackathon quickstarter orchestrator agent starts, serves a REST health endpoint returning {"status": "ok healthy"}, and a POST /message endpoint echoing input, with all Python modules importing correctly from the virtualenv. 6 分钟。

一个桌面 MCP 客户端，设计为工具统一集成，通过 Model Context Protocol (MCP) 加速 AI 采用，并实现跨供应商 LLM API 编排。

测试发现：The TUUI Electron app builds and runs: Vite produces working main/preload/renderer bundles, TypeScript type-checks pass, and the Electron app launches on the virtual display and passes both Playwright E2E tests (Document element check, Counter button click check). 4 分钟。

最具兼容性的 MCP CLI 客户端。支持：工具、资源、提示、补全、异步任务、技能、stdio 和流式 HTTP（MCP 2026-07-28 和 2025-11-25）、持久会话、带 DCR 的 OAuth 2.1、CIMD、客户端凭证和 ID-JAG、操作系统钥匙串、用于代码模式的 --json、用于工具发现的...

测试发现：The @apify/mcpc v0.7.0 project is fully installed and built with Node.js v22.14.0; all 1207 unit tests pass, all 85 e2e tests pass (3 skipped, requiring modern/legacy protocol server variants), and the CLI binary responds with version 0.7.0. 6 分钟。

MCP Server Framework 和 Tool Development library，用于为 agent 构建自定义能力。

测试发现：Arcade MCP installs from source via make install, its full pytest suite passes (4140 passed, 1 skipped), the scaffolded MCP server starts over HTTP and stdio and answers real tools/call requests (greet returned 'Hello, world!' over HTTP and 'Hello, stdio-user!' over stdio; whisper_secret returned the secret tail... 17 分钟。

160-tool MCP server，通过免费公共 API 获取实时全球情报：市场、流动性、预测市场、SEC 文件、冲突、军事、飞机群早期预警、移民与边境流动、网络、气候、新闻等。包含带日期的观测历史、29 个研究提示、引用简报、地理围栏、SSE 仪表板。MIT 许可证，无需付费 API 密钥。

测试发现：Package installed in venv, all 900 mocked tests pass, CLI responds with 48 commands, MCP server starts with 132 registered tools, cache and circuit breaker are clean. 3 分钟。

任何文件 → 干净的 Markdown 供 AI agent 使用：PDF、Word、PowerPoint、Excel、EPUB、HTML 和网页、图片（OCR）、音频和视频（元数据、字幕、转录）。一个快速的 Rust MCP 服务器和 CLI，在你的机器上运行。无需 API 密钥。

测试发现：anymd installs, builds from source with cargo release, runs all 75 TypeScript tests (0 fail, 329 assertions), and passes all Rust lib/integration tests (anymd-core 312 passed, anymd lib 79 passed, anymd-pdf 24 passed, anymd-formats 48+3 passed); the npm binary converts sample.pdf to Markdown with correct page anchors. 16 分钟。

全面的 Model Context Protocol (MCP) 服务器集合

测试发现：All 77 tests pass across 13 test files, TypeScript compiles and typechecks cleanly, catalog builder generates 8043 entries with 0 errors, profile renderer writes 8043 profiles, config-generator produces install-config JSON for any server, and the registry API server boots and serves /health (200) and /v1/servers (200). 5 分钟。

用于 Blender 的 AI 助手和 MCP 服务器：可视化聊天、场景工具、multi-agent 工作流。支持 OpenAI、ChatGPT、Claude、Gemini、Ollama、OpenCode。Blender AI 附加组件和插件，3D 副驾驶，bpy Python 脚本，MCP 集成。

测试发现：In real Blender 4.2.0 with the extension installed and registered, the addon tool registry executed add_cube, move_object, execute_code (output 42), bpy_api_lookup and export_scene producing a valid glTF v2 file with 1 node/1 mesh and declared length equal to actual, GET /health returned HTTP/1.0 200, and MCP... 84 分钟。

本地 MCP 工作笔记，支持持久会话、决策、证据与可恢复任务。

测试发现：The mcp-sequential-thinking 0.7.0 server installs from source into .venv, passes all 101 tests plus 6 subtests, answers the MCP handshake over stdio via the installed console script, records/persists/summarizes/exports/clears/imports thoughts end to end, and passes ruff and mypy --strict cleanly. 2 分钟。

开源跨 agent 记忆层，通过 MCP 为编码 agent 提供支持。兼容 Claude Code、Codex、Cursor、Windsurf、Gemini CLI、Antigravity、OpenClaw、Hermes Agent、Oh-my-Pi、Pi、Copilot、Kiro、OpenCode 和 Trae。

测试发现：Memorix builds and runs from source in this container: npm install, npm run build and lint exit 0, the full vitest suite passes 3177 tests, the CLI stores and retrieves observations and prints version 1.9.6, and both the stdio MCP smoke and the stateless HTTP MCP smoke complete successfully. 11 分钟。

Unbrowse - 逆向工程任何网站，以便你的 agent 可以直接访问其 API

测试发现：The repo builds and its tests pass; unbrowse-node serves the agent API on 127.0.0.1:7788 and the node learned and then directly called a real route (httpbin.org/json), with agent keys minted against a schema-correct in-container Secret Service mock and one test-only no-sandbox patch to passkey.rs. 80 分钟。

开源 coding agent 记忆。记录 issue、尝试、修复和决策，然后在 agent 重复已失败的方法之前发出警告。为 Claude Code、Cursor、Antigravity 和 Codex 提供原生 MCP server。100% 本地运行，无云端，无遥测。MIT 许可证。

测试发现：projectmem v0.3.3 installed in a virtual environment; the CLI tool, MCP server, test suite, and dashboard all run successfully with no errors. 2 分钟。

MCP-NixOS - 用于 NixOS 资源的模型上下文协议服务器

测试发现：All 3 NVF integration tests that were failing due to JS-chunked docs now pass, and the full test suite (495 tests) runs green; the server launches on STDIO. 8 分钟。

Git 原生的持久化内存，专为 AI coding agent 设计。实现了 Google OKF v0.2，具备亚 300µs 的内存 BM25 搜索、嵌入式 MCP 服务器和渐进式披露。将 token 膨胀削减 80%，无需任何外部数据库或依赖项。纯 Go 构建。

测试发现：The Go toolchain (v1.27.2) and bin/okf binary are built and functional: all 7 Go test packages pass, knowledge/ bundle validates with 21 concepts and 0 errors, bootstrap scaffolding creates valid starter packs, and search/show commands return correct results. 9 分钟。

一个内存层，适用于所有 AI 工具。一次性存储任何内容，然后在 Claude、ChatGPT、Cursor 或任何 MCP 客户端中调用。在 Cloudflare 的免费套餐上自托管。

测试发现：The project installs and builds under Node v24.21.0, the full test suite passes (381 files / 5671 tests), and the worker launches locally via wrangler dev answering HTTP 200 on / (dashboard) and /health with version 3.7.0 plus working MCP initialize and tools/list, limited only by Cloudflare's remote-only Vectorize... 8 分钟。

为你的 AI agent 提供持久化、集体记忆，具备去重吸收、替代谱系、语义搜索和图 UI。支持 MCP。

测试发现：Memora MCP server v0.4.6 installed in the /work/repo/.venv virtual environment, all 1088 tests pass, server starts and serves MCP memory tools (create/read/list/search/stats/delete) over streamable-http transport with TF-IDF embeddings and SQLite storage using no external credentials. 9 分钟。

一个 NestJS 模块，用于轻松创建 Model Context Protocol (MCP) 服务器，以暴露 AI 工具、资源和提示。

测试发现：Both packages build successfully (tsc), all 930 unit tests pass, and 251 e2e integration tests pass covering tools, resources, prompts, strategies, modern-era (2026-07-28) and legacy-era MCP protocol serving, guards, dynamic registration, and streamable HTTP transport. 12 分钟。

UTCP 的官方 python 实现。UTCP 是一个开放标准，允许 AI agents 直接调用任何 API，无需额外中间件。

测试发现：UTCP core library and all 10 plugin packages install in a Python 3.12 venv and pass their test suites with 569 passed, 5 skipped, 0 failures. The UtcpClient creates and closes without errors, all communication protocol types (http, sse, streamable_http, cli, mcp, text, file, websocket, tcp, udp, gql) register... 21 分钟。

自托管的 TikTok、Douyin 与 Bilibili 数据采集器和无水印视频下载器，提供异步 REST API、面向 AI agent 的 MCP 服务器、CLI 和 Web 控制台，并支持对已下载视频进行离线语音转文字。具备自愈身份池和 PostgreSQL 归档，一条 docker compose up...

测试发现：All 2960 tests pass (2450 unit + 510 integration), the CLI outputs version 5.0.1 and lists 10 commands, the API server starts and serves healthz/readyz/swagger/console root with 200, the web console SPA is built, and both the PostgreSQL+TimescaleDB and Redis backends and the Go downloader sidecar run. 48 分钟。

Agent 技能，可将自然语言、代码、Terraform/K8s、SQL、OpenAPI、AsyncAPI、Protobuf 和 GraphQL 源转换为可编辑、经过测试的 draw.io 架构图：增量同步、多视图投影、漂移差异、CI 架构测试、白板去栅格化、交互式 HTML/PPTX/Mermaid 导出。

测试发现：The drawio-skill project runs fully: 174 unit tests pass, the diagramctl CLI (12 subcommands) works end-to-end for core semantic workflows, all importers produce correct IR, the MCP server responds with 9 tools, and the shape/index/validate infrastructure is operational. Only the draw.io desktop binary (for native... 10 分钟。

一个极简但功能完善的 MCP 客户端。

测试发现：The Next.js chat app builds and runs: it serves 200 on / and /chat/[id], and /api/chats and /api/chats/[id] read and write chats/messages against a local Postgres (200/404 verified), while /api/chat streaming requires Vercel BotId OIDC and a Groq/XAI key. 37 分钟。

通过与 AI 聊天来绘制和编辑 draw.io 图表。Web 应用、桌面应用，以及用于 Claude Code、Cursor、Codex 和 VS Code 的 MCP 服务器。

测试发现：Next AI Draw.io production build completes, production server serves the web app at port 6001 returning HTTP 200 with full page content, all 809 unit tests pass, API endpoints respond correctly. 11 分钟。

开源 AI 交易操作系统，支持 agent 交易和 vibe 交易，并集成 Jev System One。研究、构建 Python 策略、回测，并在加密货币、股票和外汇中进行模拟/实盘交易。启动你自己的多租户交易 SaaS，内置用户管理、计费、支付和结算。

测试发现：QuantDinger Python API v5.4.1 is installed and fully functional, all 3050 unit tests pass, the Flask app creates and serves the health endpoint on HTTP 200, and the only skipped tests require PostgreSQL/integration Docker containers that are not available in this environment. 7 分钟。

开源 SVG 品牌库：包含 7,600+ 个品牌标志和云图标（AWS、Azure、GCP）的官方 SVG。可通过 npm、CDN、CLI、React、Vue、Svelte、Figma、VS Code、Raycast 和 MCP 搜索、复制和分发。免费、MIT、无门槛。

测试发现：Node 22.23.4 with pnpm 12.15.0, dependencies installed, 26 vitest test files (169 tests) + 48 tsx tests all pass, Next.js 16.4.2 dev server responds HTTP 200 on / via port 3333. 3 分钟。

GitHub 的官方 MCP 服务器

测试发现：The GitHub MCP Server builds from source, all unit tests pass, and the HTTP server responds to MCP initialize requests with HTTP 200 and a complete capabilities object. 7 分钟。

MCP 服务器为 Cursor 等 AI coding agents 提供 Figma 布局信息。

测试发现：The Figma MCP server installs, builds, type-checks, passes all 250 tests, and runs an HTTP MCP server that accepts initialize and tools/list requests on the configured port. 4 分钟。

HexStrike AI MCP Agents 是一个高级 MCP 服务器，允许 AI agent（如 Claude、GPT、Copilot 等）自主运行 150 多个网络安全工具，用于自动化渗透测试、漏洞发现、漏洞赏金自动化和安全研究。它无缝地将 LLM 与现实世界的进攻性安全能力连接起来。

测试发现：HexStrike AI MCP Server v6.0 starts on any port, serves Flask API with 30+ endpoints including command execution, file management, intelligence analysis, process management, caching, and telemetry. FastMCP client connects and registers 16+ security tools. 34 分钟。

Windows 上的计算机使用 MCP 服务器

测试发现：285 tests passing, 321 skipped (Windows-only); CLI entry point responds to --help; MCP server with stdio/SSE/streamable-http transport available. 64 分钟。

BioMCP：生物医学模型上下文协议

测试发现：BioMCP 0.9.1 builds, installs to PATH, runs CLI queries against 51 live biomedical APIs, serves MCP over HTTP (health endpoint returns 200), and passes 3909 Rust tests plus 1195 Python contract tests. 55 分钟。

Gentle-AI 配置你已经在使用的 AI coding agents：Claude Code、Cursor、OpenCode、Codex、Pi 等。可选择持久记忆、Organic-Driven Development、精选技能、MCP 服务器、角色以及可选的有界审查。开源，无 agent 锁定。

测试发现：gentle-ai 2.0.0 builds from source, prints version (2.0.0-20260906031619-2c25e878eea4) and help text, all 53 Go test packages pass. 17 分钟。

mcp-language-server 为支持 MCP 的客户端提供语义工具，例如获取定义、引用、重命名和诊断。

测试发现：The MCP language server builds and runs; it initializes as an MCP server, connects to gopls, pyright, typescript-language-server, and rust-analyzer via stdio, and exposes 6 tools (definition, references, diagnostics, hover, rename_symbol, edit_file) that work correctly with real language servers across Go, Python... 62 分钟。

面向 xiaohongshu.com 的 MCP

测试发现：The HTTP server starts on port 18060, the /health endpoint returns 200, the built-in Chromium browser launches via go-rod, and the /api/v1/feeds/list endpoint returns real feed data from xiaohongshu.com. All 10 Go test packages pass (0 failures). The browser extraction bug is fixed by using system tar instead of the... 18 分钟。

官方 Firecrawl MCP 服务器，面向 AI agent 的上下文 API：从任何 MCP 客户端搜索、抓取和与 Web 交互，以及解析、爬取、映射、Agent 和监控。

测试发现：Firecrawl MCP Server v3.24.1 builds, installs, and starts successfully with 25 tools in stdio mode; 81 of 84 tests pass. 24 分钟。

为本地开发批量转发 Kubernetes 服务的端口。

测试发现：The kubefwd Go project builds, all 24 test packages pass with race detection and coverage profiling, and the binary reports its version and help text successfully. 7 分钟。

用于 Grafana 的 MCP 服务器

测试发现：Go toolchain installed from source, binary builds cleanly, all 973 unit tests pass, and the MCP server starts and responds to tool discovery requests. 7 分钟。

AI 上下文的 ripgrep：一个零依赖的 C++23 CLI + MCP 服务器，用于编码 agent。无需阅读仓库即可找到所需内容，然后检查你构建的是否符合预期：blast radius、要运行的测试、质量差异。签名比函数体少 74.7% 的字节；每个猜测都被标记，每个损失都公布。带着地图出发。

测试发现：Build succeeds (plain dev build/ and Release build-install/), binary parses the repo source tree and emits deterministic minified XML with ranked symbols via Personalized PageRank, installed to /home/runner/.local/bin/ripwire with skills and hooks, --version/--help work, Python indexing confirms correct function... 21 分钟。

Google Research 的 PaperBanana 的开源实现与扩展，用于自动生成学术图表、示意图和研究可视化，并扩展到幻灯片生成等新领域。

测试发现：PaperBanana v0.3.0 is installed and its full test suite (901 tests) passes without errors. The CLI responds to all commands. The only missing pieces are external API keys (GOOGLE_API_KEY, OPENAI_API_KEY) which are expected for a fresh install. 16 分钟。

用于 Postgres 技能和文档的 MCP 服务器和 Claude 插件。帮助 AI 编码工具生成更好的 PostgreSQL 代码。

测试发现：bun i installs 356 packages, tsc -p tsconfig.build.json compiles to dist/, bun test runs 66 passing tests, bun run validate-skills validates all 10 skills, and ./check passes (skipping Python checks for missing uv). 6 分钟。

面向自主 AI agent 的安全*运行时。策略源自简明英语章程。

测试发现：IronCurtain 0.14.0 installed from source, builds cleanly with tsc, and runs in builtin agent mode: it accepts a task via the CLI, connects to a configurable LLM provider at ANTHROPIC_BASE_URL, and returns the model response. The test suite passes 7783 of 7785 tests (2 pre-existing failures: one UX test mismatch and... 17 分钟。

Ghidra MCP Server：200+ 个 MCP 工具，用于 AI 驱动的逆向工程。包含 GUI 插件和无头服务器，支持懒加载工具、约定强制、批量操作、Ghidra Server 集成以及 Docker 部署。

测试发现：Python bridge-mcp-ghidra 7.0.0 installs, all 552 non-slow unit tests pass, MCP server starts on stdio/sse/streamable-http transports and responds to initialize and tools/list with 8 static bridge tools; integration tests require a live Ghidra instance not available in this container. 5 分钟。

社区插件，可使用你选择的任何 LLM 控制 Blender 3D。与官方 Blender Foundation 无关。

测试发现：The blender-mcp package is fully installed under uv, all 113 tests pass, and the MCP server launches correctly over stdio. 4 分钟。

一个强大的 MCP 编码工具包，提供语义检索和编辑能力，是你的 agent 的 IDE。

测试发现：Serena is installed, initialized, and the MCP server starts successfully with all 708 tests passing. 12 分钟。

ToolHive 是一个企业级平台，用于运行和管理 Model Context Protocol (MCP) 服务器。

测试发现：ToolHive v0.51.3 builds from source with the standalone Go 1.27.1 compiler. The thv binary starts, prints its help page, and outputs version/build information. Unit tests pass at 198/201 with the 3 failures all due to missing Docker or system keyring in this container. 55 分钟。

适用于你的 monorepo 的终极 RAG。借助 AI 和知识图谱的力量，查询、理解并编辑多语言代码库。

测试发现：code-graph-rag 0.0.985 installed and building from source; CLI responds with version 0.0.985; 185 non-integration tests pass, 45 skip due to missing optional grammars; Docker+Memgraph unavailable without Docker. 55 分钟。

DaVinci Resolve Studio 的 MCP 服务器集成

测试发现：Python compound MCP server starts and answers MCP initialize/tools/list returning 37 tools. Python test suite: 3554 offline tests pass. Node.js resolve-advanced test suite: 913/961 pass (remaining failures from Node.js version gap v18 vs >=20.9). npm install and venv install complete. DaVinci Resolve not available in... 22 分钟。

官方、AWS 支持的 MCP 服务器、技能和插件，帮助 AI agents 在 AWS 上构建

测试发现：All repository validation scripts (manifest, spec conformance, skill sync, markdown lint) pass. 275 embedded skill tests across agent-advisor, cloudwatch-omni, and openclaw-bridge suites all pass. 3 pre-existing test gaps exist in agents-pay (httpx isolation, exit code expectations, Typescript build requirement). 4 分钟。

一个聚合来自不同 Web 搜索服务的结果的元搜索库

测试发现：The ddgs package installs, lints, and formats cleanly; the CLI and Python API produce real results for text, images, news, and extract; videos and books tests fail because DuckDuckGo's v.js endpoint and Anna's Archive return 403 from this container. 12 分钟。

用于从 arXiv、PubMed、bioRxiv 等多个来源搜索和下载学术论文的 MCP、CLI、Skills。

测试发现：Package installs, imports, CLI search/results, and MCP server handshake all work. Tests pass modulo transient rate limits from external API servers. 9 分钟。

Apify MCP 服务器使你的 AI agents 能够使用 Apify Store 上提供的数千个现成的抓取器、爬虫和自动化工具，从社交媒体、搜索引擎、地图、电商网站或任何其他网站提取数据。

测试发现：Dependencies install, type-check, lint, format, unit tests, and build all pass; the dev server starts on port 3001 and responds to MCP initialize with HTTP 200 and valid server capabilities. 7 分钟。

ChatGPT 负责思考。Codex 负责执行。将 ChatGPT 作为规划大脑，同时保留 Codex harness。

测试发现：The project installs, builds, and passes all 164 tests; the bridge server starts on localhost port 48765 and responds to health checks with status ok. 5 分钟。

源代码复制粘贴检测器。支持 220+ 种语言，Rust 引擎，SARIF/HTML/badge 报告器，GitHub Action，面向 AI agent 的 MCP 服务器。

测试发现：jscpd 5.1.1 builds from source via cargo, installs via npm as a prebuilt binary, and runs duplication detection on 224 language formats with 15 reporters; the full Rust test suite passes. 9 分钟。

面向 AWS 的开源 MCP 服务器

测试发现：58 AWS MCP server projects install and build with uv; ~15,000 unit tests pass across 50+ servers; the aws-documentation-mcp-server launches, responds to MCP initialize over stdio, and makes real API calls to AWS Documentation Search and docs.aws.amazon.com. 35 分钟。

编码、构建和评估 agents：出色的模型与技能/MCP/ACP/A2A 支持。

测试发现：All 8730 unit tests pass, CLI boots and responds, playback LLM agent runs end-to-end, format/lint/typecheck all pass. 30 分钟。

用于构建、测试和部署生产就绪 MCP 服务器和 AI 原生应用的全栈 TypeScript 框架。

测试发现：All three NitroStack packages (core, cli, widgets) build and all 892 tests pass on Node.js 18.19.1. The CLI binary responds to --help. The core server starts over stdio, initializes, lists tools, and executes tools successfully. 75 分钟。

面向开发者的自托管 API 管理

测试发现：Fusio v8.8.7 API server running on PHP 8.4 with SQLite at http://127.0.0.1:8084; CLI commands work; backend admin UI at /apps/fusio/ returns 200; system/health endpoint returns healthy: true. 18 分钟。

工程师及其 AI agent 的团队记忆。存在于你的仓库中。通过 Git 共享。

测试发现：MEX 0.8.2 built, typechecked, and tested with Node 24.0.0 (downloaded). CLI version 0.8.2 confirmed. Test suite passes 4288/4317 tests with 6 pre-existing failures (4 timeouts on constrained hardware, 2 test bugs treating directories as files). npm install, build, and typecheck all succeed. FTS5 full-text search... 26 分钟。

一个拥有自己计算机的 AI 同事。自我进化、持久记忆、MCP 服务器、安全凭证收集、电子邮件身份。基于 Claude Agent SDK 构建。

测试发现：2639 tests pass, the HTTP server starts on port 3100 and returns 302 on GET /, and the CLI phantom doctor command reports the project configuration is valid. 34 分钟。

一个用于将 Notion 连接到 MCP 兼容客户端的 Model Context Protocol 服务器

测试发现：Install and build succeed, all 142 vitest tests pass, and the MCP server starts and responds to MCP protocol initialize/tools/list requests. 2 分钟。

Query MCP 通过聊天界面实现对 Supabase 的端到端管理：支持读写查询执行、管理 API、自动迁移版本控制、访问日志等。

测试发现：The Supabase MCP server installs cleanly, all 191 tests pass, and the server binary starts and answers the MCP initialize/tools-list handshake with 12 registered tools. 14 分钟。

TypeScript AI SDK 和 agent 框架，零依赖。构建生产级 AI agent 和多 agent 集群：持久执行、长期记忆、混合 RAG、MCP 客户端/服务器、A2A、人工审批、CodeAct 沙箱、实时语音。一个流式 API，29...

测试发现：Both @deuz-sdk/core and @deuz-sdk/react build from source and pass their full test suites (2635+39 tests passing) using MSW-mocked HTTP providers, requiring no real API credentials. 7 分钟。

面向 AI 编码 CLI 的多智能体编排：Claude Code、Kiro、Codex 等，在隔离的 tmux 会话中协调。

测试发现：CAO server starts, responds 200 on /health and /docs, and passes all unit tests (2986+ passed) except for 4 pre-existing test-isolation wiki_lint failures and Node version-gated web UI build. 43 分钟。

开源、自托管的指标、日志和数据库告警，包含内置 AI agent 和 MCP server。

测试发现：n9e server builds, starts on port 17000 with SQLite+embedded TSDB, serves SPA and REST API, authenticates root user login, responds to Prometheus queries, and exposes a built-in MCP endpoint with 30+ tools over Streamable HTTP transport. 13 分钟。

这是一个用于 Claude 的 MCP 服务器，为其提供终端控制、文件系统搜索和 diff 文件编辑能力。

测试发现：The @wonderwhy-er/desktop-commander MCP server v0.2.48 builds, installs, starts, responds to MCP protocol initialize, and passes its full test suite including real file operations, Excel, PDF, REPL, and process management tests. 9 分钟。

LeanCTX 是 AI 系统的上下文网关。选择任务相关的上下文，应用支持的控件，连接本地知识并检查记录的证据。开源引擎：CLI、MCP、钩子和支持的代理路径。

测试发现：lean-ctx 3.10.1 release binary is built, installed at ~/.cargo/bin/lean-ctx, and fully functional with all CLI commands working. 65 分钟。

面向 coding agents 的 Chrome DevTools

测试发现：The project builds, all tests pass (including e2e), and both CLI and MCP server start successfully using the puppeteer-installed Chrome browser. 51 分钟。

为开发者和 AI agent 提供本地代码智能。在持久化图中连接代码、文档和基础设施，并提供 CBM Lens、MCP 工具、agent 集成以及变更感知的测试选择。

测试发现：codebase-memory-mcp v0.10.8 is installed, the MCP server is configured in Codex CLI config.toml with hooks, and the repository is indexed with a knowledge graph of 21,005 nodes and 119,361 edges, all 15 MCP tools verified working. 15 分钟。

数据访问与运营智能的交汇之处

测试发现：WhoDB backend and CLI binaries build and run, all core unit tests pass, and frontend compiles successfully. 30 分钟。

一个 AI Gateway、注册表和代理，位于任何 MCP、A2A 或 REST/gRPC API 之前，暴露统一端点，提供集中发现、护栏和管理。优化 Agent 和 Tool 调用，并支持插件。

测试发现：ContextForge gateway installed in .venv, server launches and responds on HTTP, 592+ unit tests pass against real SQLite and auth infrastructure. 32 分钟。

使用 agents 对任何内容进行逆向工程，从应用行为到原生二进制文件。

测试发现：npm ci installs 281 packages, tsc builds cleanly to dist/, and the full vitest suite runs 353 test files (1756 tests) with zero failures on Node 22.19.0. 7 分钟。

Android 全栈设备控制平台：WebRTC/H.264 远程桌面、UI/OCR/图像匹配自动化、一键 MITM、内置 Frida、代理/VPN/frp/P2P 网络、MCP/Agent、160+ 个 API，专为多设备集群和工程化部署而设计。

测试发现：lamda Python client v10.8 installs, all modules import, and the gRPC protocol stack works end-to-end via a mocked server. 7 分钟。

WAHA - WhatsApp HTTP API（REST API），一键即可配置！支持多种引擎：WEBJS（基于浏览器）、NOWEB（websocket nodejs）、GOWS（websocket go）、WPP（浏览器）

测试发现：The project builds, all 308 unit tests pass across 31 suites, and the server starts and responds to HTTP API calls (GET /api/sessions returns [] with 200 status, logged 'Nest application successfully started' and 'WhatsApp HTTP API is running on: http://[::1]:3000'). 62 分钟。

它类似于 v0，但就在你的 Cursor / Claude Code / Windsurf 中：搜索 10,000+ 个 React/Tailwind 组件，使用 AI 生成新的 UI，并发布你自己的组件，全部直接在编辑器内完成。Magic MCP 现在是第 21 个...

测试发现：The MCP compatibility proxy builds, installs, and runs as a stdio-to-HTTP proxy, correctly handling initialize, tools/list, auth-failure, and auth-latch scenarios against a local mock server. 3 分钟。

用于 Atlassian 工具（Confluence、Jira）的 MCP server

测试发现：Project installs with uv sync, 3938 unit tests pass, and the MCP server binary starts via SSE transport and prints its version banner. 7 分钟。

Klavis AI: MCP 集成平台，让 AI agent 在任何规模下都能可靠地使用工具。

测试发现：The strata-mcp package installs and imports cleanly. The CLI parses all commands (add, remove, list, enable, disable, auth, run, tool). Stdio mode starts without errors. HTTP server mode starts on port 8080, and the StreamableHTTP MCP endpoint /mcp/ responds 200 to tools/list requests, returning the Strata tool... 15 分钟。

Zotero MCP：通过 Model Context Protocol 将你的 Zotero 研究文库与 Claude 和其他 AI 助手连接起来，用于讨论论文、获取摘要、分析引用等。

测试发现：Zotero MCP v0.13.3 installed in a Python venv, all modules load, CLI responds to help/version commands, and pytest reports 3785 passed, 0 failed, 1 skipped across the non-live test suite after fixing one environment-leak bug in a test fixture. 11 分钟。

Exa MCP，用于网页搜索和网页爬取！

测试发现：The Exa MCP server builds, all 154 unit tests pass, and the stdio binary starts and correctly serves the JSON-RPC protocol with web_search_exa and web_fetch_exa tools declared. 3 分钟。

ACI.dev 是一个开源工具调用平台，通过直接函数调用或统一的 MCP 服务器，将 600 多个工具接入任何 agentic IDE 或自定义 AI agent。它是 VibeOps 的发源地。

测试发现：The ACI backend services (FastAPI server, PostgreSQL with pgvector, AWS KMS mock, Propelauth auth mock) and frontend (Next.js dev portal) install and test successfully in this container without Docker. 159 backend tests pass covering health checks, apps CRUD, functions execution, linked accounts, projects, app... 37 分钟。

通过 Obsidian rest API 社区插件与 Obsidian 交互的 MCP 服务器

测试发现：All 81 unit tests pass, the MCP server responds to initialize and tools/list over stdio, and the Obsidian client makes successful HTTP calls against a mock REST API server. 4 分钟。

将 Claude Code 或 Codex 转变为最智能的 Deep Research Agent。收集、搜索并综合网络研究，形成一个持久化、可搜索的、自我构建的 wiki。托管 API + MCP: hyperresearch.ai

测试发现：All 1719 pytest tests pass, CLI reports version 0.12.0, vault initialization and web server launch succeed, and both `hyperresearch` and `hpr` entry points are functional. 3 分钟。

开源 AI 助手生态系统，支持 MCP 集成、多模态工作流、IoT 支持以及跨平台语音交互。

测试发现：80 test cases pass in the project test suite; CLI help and CLI mode launch successfully. 16 分钟。

Google Flights MCP、CLI 和 Python 库

测试发现：fli installs cleanly with uv, all 1404 offline tests pass (29 skipped in container config), the CLI parses and displays airport data, the MCP HTTP server starts and serves /health (200) and /mcp/ (307). 8 分钟。

使用任何 LLM 控制 Ableton Live：通过 MCP 创建轨道、编排片段和创作音乐。

测试发现：Ableton MCP v1.4.5 is installed in a Python 3.12 venv, all 49 unit tests pass, and the MCP server starts and responds to MCP protocol messages over stdio transport against a mock Ableton backend. 5 分钟。

用于 AI agents、MCP 服务器和 agent 技能的安全扫描器。

测试发现：Project installs with `uv sync`, CLI prints help and version, and the full test suite passes 2293/2326 collected tests. 34 分钟。

技能、MCP 服务器、自定义 Agents，以及用于 SDK 为 Coding Agents 提供基础的 Agents.md

测试发现：The test harness installs and evaluates all 136 skills with mock mode, scoring 96.3% pass rate across 1227 scenarios; vitest unit tests are blocked by a missing native binary in the rolldown dependency. 8 分钟。

KiCAD MCP 是一个 Model Context Protocol (MCP) 实现，它使像 Claude 这样的大型语言模型 (LLM) 能够直接与 KiCAD 交互，用于印刷电路板设计。

测试发现：Installation, TypeScript build, and full test suite (112 TS + 2611 Python tests) all pass. Server registers all 244 tools at startup but requires KiCAD's pcbnew module for the Python backend, which is unavailable in this container and is mocked by the test suite's conftest.py. 5 分钟。

AI Agent 优先的 JS 逆向 MCP 服务器：有头 Chrome 调试、断点、网络/WebSocket 分析、Patchright 反检测，可选 CloakBrowser。

测试发现：The js-reverse-mcp v4.0.5 TypeScript project compiles to JavaScript, its full test suite of 114 tests passes, and the MCP server binary responds to JSON-RPC initialize on stdin/stdout. A Node.js v22.12.0 upgrade was applied (portable) and the cloakbrowser optional dependency was installed. Google Chrome is not... 4 分钟。

一个强大的 Model Context Protocol (MCP) 服务器，提供公共 Web 访问的一站式解决方案。

测试发现：The Bright Data MCP server installs and all 21 unit tests pass under Node.js v22.12.0; the server starts listening when API_TOKEN is set. 4 分钟。

自托管的 MCP 网关和控制平面，用于连接、控制和运行 MCP 服务器。

测试发现：Backend compiles (tsc exits 0), frontend builds (1902 modules, vite exits 0), 99+ test suites pass (919+ tests, no failures), app server starts on :3000 and responds with 200 on /health, /, and /login; MCP servers report 'degraded' because uvx/npx for fetch/amap/playwright/slack are not installed in this container... 76 分钟。

Perplexity API Platform 的官方 MCP 服务器实现

测试发现：npm install builds successfully, all 90 tests pass, HTTP server starts and serves the /health endpoint and MCP tools/list endpoint correctly. 2 分钟。

一句话 → 你的 AI 专家一人公司 → 几分钟内交付完整方案。276 CN + 184 EN + 另外 5 种语言（ko/ru/pt-BR/id/ar）· 零代码 YAML · 自动验证的验收 · Web Studio / 桌面 / Docker · 15 个 LLM 提供商（11 个无需密钥）。

测试发现：Agency Orchestrator 0.19.2 installed and built successfully on Node.js v22, CLI answers version/validate/roles commands, web Studio API serves health/workflows/roles endpoints, and the majority of unit tests pass. 49 分钟。

AssetOpsBench：Industry 4.0：一个统一的 benchmark 和框架，用于构建、编排和评估面向 Industry 4.0 资产运营与维护的特定领域 AI agents，包含 460+ 个场景、5 个专家 agent（IoT、FMSR、TSFM、Work Order 等），以及基于 MCP 的多...

测试发现：Project installs, builds, and 675 of 702 tests pass; 27 integration tests skip due to missing CouchDB/LLM credentials; all CLI entry points import and display help. 25 分钟。

用于 Kubernetes 和 OpenShift 的 Model Context Protocol (MCP) 服务器

测试发现：Go 1.26.4 installed, kubernetes-mcp-server binary built (102MB), all 39 test packages pass, binary runs and prints help/version, runtime requires a Kubernetes kubeconfig. 29 分钟。

一个 Model Context Protocol 服务器，提供对 MySQL 数据库的只读访问。该服务器使 LLM 能够检查数据库模式并执行只读查询。

测试发现：151 unit tests pass; MCP server launches in both stdio and remote HTTP modes, responding to initialize requests with proper capabilities. 17 分钟。

用于 AI agents 的本地辅助 BOSS Zhipin CLI：搜索、福利筛选、候选名单、JSON 信封输出；默认低风险且合规。

测试发现：The project installs and builds under uv, the whole offline test suite passes 2084/2084, ruff and mypy pass, the boss CLI and the MCP server start and answer, and the CLI's status, detail, search, welfare-filter and wizard commands return correct ok:true JSON envelopes against a local mock of the zhipin API that... 14 分钟。

Awesome MCP Servers 目录

测试发现：Next.js 15.0.4 app builds cleanly and serves all routes including homepage, categories, privacy-policy, terms-of-service, sitemap, favicon, and API endpoints on port 8051. 6 分钟。

完全免费、私密、基于 UI 的技术文档 MCP 服务器。专为程序员和软件开发者设计。可轻松集成到 Cursor、Windsurf、Cline、Roo Code、Claude Desktop App。

测试发现：Backend (FastAPI on :24125) and frontend (Next.js on :3001) both start and respond 200. The discover/crawl/crawl-status API endpoints successfully process a job from initialization through discovery_complete to completed, writing consolidated markdown and metadata files with crawl results including HTTP status codes. 23 分钟。

面向 SEO 的 Google Search Console Insights 与 Claude AI 集成

测试发现：All 52 unit tests pass with mocked GSC API. The server loads, registers 21 MCP tools, completes MCP initialization handshake, responds to tools/list and tools/call via stdio transport, and serves 200 on the SSE endpoint. 3 分钟。

ADR 通过可观测性、安全 benchmarking 和威胁检测来保护企业 AI agents。已在 Uber 部署。

测试发现：All three Python packages (adr-benchmark, adr-sensor, adr-discovery) install and pass their test suites (802 total tests, 0 failures). The llamaFirewall detector pipeline runs against packed benchmark data with a mock API key, scoring conversations and writing analysis results. CLI entry points for Discovery and... 11 分钟。

用于 GitLab 的 MCP 服务器：通过 GitLab REST/GraphQL API 提供 277 个工具，涵盖项目、MR、issue、流水线、wiki 等。

测试发现：The @zereight/mcp-gitlab MCP server version 2.3.2 installs, builds, and passes its full mock test suite (all tests pass) using a built-in mock GitLab server that replaces the real GitLab API. 22 分钟。

用于 Azure DevOps 的 MCP 服务器，将 Azure DevOps 的强大功能直接带给你的 agents。

测试发现：Project builds and all 1349 tests pass with Node.js v20.20.2. The MCP server binary reports version 2.10.0 and responds to --help with all options (organization, domains, authentication, tenant). The dist/ directory contains compiled JS output. 3 分钟。

由 Telethon 驱动的 Telegram MCP 服务器，让 MCP 客户端能够读取聊天、管理群组，并发送/修改消息、媒体、联系人和设置。

测试发现：The project installed via uv sync with all 70 dependencies, the 791-test suite passes, the session generator and migration scripts produce valid help output, and the MCP server imports and initializes its Telegram client infrastructure correctly when env vars are set (real Telegram credentials required for full... 4 分钟。

最强大的 MCP Slack 服务器，无需权限要求，支持应用、GovSlack、私信、群私信以及智能历史获取逻辑。

测试发现：slack-mcp-server builds, unit tests pass (30+ tests), and the server starts in demo mode serving SSE and HTTP MCP endpoints with all tools registered. 15 分钟。

Dive 是一个开源的 MCP Host 桌面应用，可与任何支持函数调用能力的 LLM 无缝集成。

测试发现：Installed with Node v22.14.0 and uv 0.12.23; `npm run build:electron` and `npm run check` exit 0; 287 mcp-host pytest pass; Electron app launches on Xvfb display with host reporting state UP and its HTTP API answering 200 on three endpoints. 27 分钟。

生产级 MCP 服务器，为 Claude 提供 27 个安全情报工具，涵盖 21 个 API：CVE 查询、EPSS 评分、CISA KEV、MITRE ATT&CK、Shodan、VirusTotal 等。

测试发现：30/30 pytest tests pass, server starts in both stdio and streamable-http transport, NVD rate-limit warning confirmed expected without API key. 2 分钟。

用于 Kubernetes 管理命令的 MCP Server

测试发现：26 unit test files pass (297 tests), the build compiles cleanly, the MCP server responds to initialize and tool calls, and all tool interactions are handled through a mock kubectl since no real cluster is available. 14 分钟。

官方的 MiniMax Model Context Protocol (MCP) 服务器，支持与强大的文本转语音、图像生成和视频生成 API 进行交互。

测试发现：Package installed in a venv, all 9 pre-existing tests pass (3 server + 6 utils), and the MCP server binary launches and waits for stdio input. 2 分钟。

一个通过 DuckDuckGo 提供网页搜索能力的 Model Context Protocol (MCP) 服务器，并具备内容获取与解析的附加功能。

测试发现：All 136 tests pass (128 unit + 8 e2e), ruff lint passes, and the server starts on streamable-http transport and prints its configuration banner and listen address. 1 分钟。

Terraform MCP Server 提供与 Terraform 生态系统的无缝集成，为基础设施即代码（IaC）开发提供高级自动化与交互能力。

测试发现：Build compiles, binary runs, stdio MCP mode responds correctly to JSON-RPC, streamable-http mode starts and binds port. 16/16 test packages pass, 1 e2e package fails due to missing Docker in environment. 4 分钟。

使用 MCP 和 ROS 将 Claude、GPT 等 AI 模型与机器人连接。

测试发现：ros-mcp 3.1.2 installed in /tmp/venv, CLI responds with usage, MCP server runs over stdio/http with full tool/resource/prompt registry, 12/12 unit tests pass, tools verified against mock rosbridge WebSocket server. 60 分钟。

使用原生 Rust 编写的开源 AI agent 框架，从单个二进制文件提供 GUI、CLI、无头模式和 webapp。支持多提供商、MCP、技能、插件和 agent 团队。

测试发现：Rust 1.99.0, Node 22.14.0, and pnpm 10.34.5 installed. thclaws-core 0.138.0 compiles successfully (non-GUI). Both thclaws-cli and thclaws binaries respond with correct version. Frontend dist built. 2669/2676 unit tests pass; 7 pre-existing thclaws_gateway tests fail with PoisonError from reentrant env locking. 16 分钟。

Atlassian 的官方远程 MCP 服务器。使用 OAuth 2.1 或 API 令牌，安全地将 Jira、Confluence、Jira Service Management、Bitbucket 和 Compass 连接到 Claude、ChatGPT、Cursor、VS Code 及其他 AI 工具。

测试发现：The repo's install and validation work: npm ci succeeds, npm run validate passes, and the full node test suite passes 17/17, with a schema-correct local mock MCP server confirming the documented streamable-HTTP handshake (initialize 202 + tools/list) that the packaged client configs point to. 4 分钟。

通过统一的基于 SQL 的框架，为人类和 AI agent 查询、配置和操作 Cloud、SaaS、API 以及 Model Context Protocol (MCP) 资源。

测试发现：Stackql binary v1.1.1 builds, runs, and passes 512/554 robot functional tests against bundled Flask mocks (42 failures due to missing psql/background-srv infra, not code defects); all 53 Go unit test packages pass. 38 分钟。

coding agent 的实时消息

测试发现：The project installs and builds cleanly on Node 22.17.0 with pnpm 10.33.0, its CI commands all pass, and its CLI and MCP server ran through a complete two-agent claim, prompt, reply, transfer, and review workflow in a scratch workspace. 4 分钟。

一个中间件，用于提供兼容 OpenAI 的端点，可调用 MCP 工具。

测试发现：uv sync installed all 70 dependencies and built the project, the server starts and responds to health (200), MCP tools/prompts/resources (200 each), docs (200), and SSE endpoint (200). The /v1/models endpoint returns 500 because no inference server is configured. 6 分钟。

简洁、面向 LLM 优化的 Reddit MCP 服务器。浏览帖子、搜索内容、分析用户。无冗余，只有 Reddit 数据。

测试发现：The app installs and builds cleanly, its unit and integration suites both exit 0 (45/45 and 68 passed/39 skipped), the CLI reports v1.1.14, the HTTP MCP server answers 200 and serves tool results, and the credentialed browse/search/post/user paths work end-to-end against a schema-correct mock through the real binary. 22 分钟。

开源、自托管的 Composio、Nango、Merge.dev 和 Arcade.dev 替代方案，用于 AI agent 集成。可通过 MCP 或 REST 运行 GitHub、Slack、Hacker News 以及自定义 HTTP 操作。

测试发现：Server starts on port 8000, responds to /health with 200, lists 236 providers via /v1/providers, exposes 8,024 actions via /v1/actions, and all 134 in-process pytest tests pass; web console builds with Next.js 16. 9 分钟。

MCP Server for Adobe After Effects。支持通过 Model Context Protocol 使用 ExtendScript 远程控制合成、文本、形状、固态层和属性。

测试发现：npm install succeeds, esbuild build produces build/index.js (27.1KB), and the MCP stdio server starts and responds to all 13 registered tool calls and 3 prompts over JSON-RPC protocol. 2 分钟。

利用此 TUI MCP Client for Ollama 发挥本地 LLM 的强大功能。具备所有核心 MCP 原语（工具、提示词、资源）、agent 模式、多服务器、模型切换、流式响应、人在回路、思考模式、模型参数配置、系统提示词以及保存的偏好设置。

测试发现：The ollmcp CLI installs from source, builds and renders help, and its full test suite passes 305/305 after a one-file Rich word-wrap fix; MCP server add/list/remove works, but live LLM chat cannot be exercised because no Ollama server is available. 6 分钟。

JADX-AI 插件的 MCP 服务器

测试发现：MCP server starts, accepts initialize handshake, registers 34 tools, and proxies all tool calls to a JADX plugin backend via HTTP, returning proper SSE responses. 7 分钟。

Adobe Premiere Pro MCP。通过 MCP 为 Codex、Claude 及其他 MCP 客户端提供 AI 驱动的视频编辑工具。

测试发现：npm install and tsc build succeed, all 653 tests pass, the MCP server starts and responds to tools/list, tools/call, resources/list, and prompts/list over stdio with a 283-tool catalog, 13 resources, and 10 prompts. 14 分钟。

适用于 Adobe Photoshop 的 MCP。118 个工具。从 Cursor、Claude 或自定义 LLM 进行控制。

测试发现：Project installs and builds cleanly; the MCP server starts and answers real JSON-RPC with 128 tools and 24 prompts, the standalone UI serves on 127.0.0.1 with 200 responses on / and token-protected /api/health, the 35-suite unit suite passes 254/254 (+1 skip) under the container's Node 18, and the only unimplemented... 34 分钟。

通过 MCP 读取、写入、搜索并精准编辑 Obsidian 仓库中的笔记、标签和 frontmatter。支持 STDIO 或 Streamable HTTP。

测试发现：Dependencies installed, project builds, all 2608 tests pass, and the server starts via stdio with 14 tools and 3 resources registered. 2 分钟。

一个简单的 CLI，用于运行 LLM 提示并实现 MCP 客户端。

测试发现：Package installs, CLI entry point works, and the agent executes queries against an LLM endpoint with streaming, returning expected output. 6 分钟。

本地优先、自托管的 AI agent 运行时和 MCP 桥接，具备沙箱会话、内存、凭据、审计/回放以及本地 Console。

测试发现：The SandBase Harness runtime starts on port 3000 with healthy API/Dashboard endpoints, its full test suite passes, and a CLI chat turn against a local mock OpenAI-compatible provider returns 'Hello from the mock provider!' and records the turn with stop_reason 'stop' and 5/6 tokens in the session event log. 41 分钟。

一个使用任何 LLM 提供商对任何数据进行深度研究的自主 agent。

测试发现：The GPT Researcher package is installed in a Python venv. The FastAPI backend server launches, listens on 0.0.0.0:8000, and serves the GPT Researcher frontend (HTTP 200). The CLI help works. 410 of 411 tests pass under --forked isolation; 2 integration tests require real OpenAI credentials and are skipped. No code was... 23 分钟。

本项目为xiaozhi-esp32提供后端服务，帮助您快速搭建ESP32设备控制服务器。Backend service for xiaozhi-esp32, helps you quickly build an ESP32 device control server.

测试发现：Python xiaozhi-server module starts successfully with WebSocket (port 8000) and HTTP (port 8003) servers, 59/59 Python tests pass, 53/53 web vitest tests pass, 9/9 contract tests pass, OTA endpoint returns HTTP 200; Java manager-api module and manager-mobile could not be tested due to missing JDK/Maven in container. 12 分钟。

开源 AI 安全运营中心：告警融合、LLM-agent 分类、MITRE ATT&CK 调查，以及记录每个 agent 步骤的可回放决策账本。可自托管，无需 API 密钥即可运行，MIT 许可证。附带一个适用于 Claude、Cursor 和 Continue 的 MCP 服务器。

测试发现：All 7 Python packages install and their 249 tests pass. aisoc-sandbox runs a 4-step investigation offline in 1.3s. aisoc-lite CLI triages 200 demo alerts and translates Sigma rules. Docker is unavailable so the full production stack (make up -> make smoke) cannot run. 11 分钟。

用于 Ghidra 的 MCP 服务器

测试发现：The ghidra-mcp Python package (v2.0.0) is installed in a uv-managed .venv, all 19 mock-based tests pass, the MCP server starts via stdio and streamable-http transports and responds with proper tool listings; the Ghidra Java extension, headless tests, and any end-to-end reverse-engineering workflow require Ghidra... 3 分钟。

本地部署的对话式 RAG，支持可配置容器

测试发现：Chat frontend builds and test passes, Electron app launches on Xvfb, MCP server responds to MCP initialize/tools/list protocol, indexer Python module passes standalone embedding, indexing, search, user-isolation, and removal tests with in-memory Qdrant; full Docker stack (Qdrant, Ollama) not available in container so... 37 分钟。

用于与 iOS 模拟器交互的 MCP 服务器

测试发现：The ios-simulator-mcp server builds, starts, and responds to all 17 MCP tools correctly on Linux with mocked external dependencies. Real functionality requires macOS with Xcode, iOS Simulators, and Facebook IDB. 5 分钟。

通过 MCP 连接 IDA Pro 与语言模型的 AI 逆向工程助手。

测试发现：ida-pro-mcp package installs via uv sync, 255 standalone tests pass, the MCP server responds to initialize/ping over stdio and HTTP, and idalib-mcp/ida-mcp-test are blocked by missing IDA Pro libidalib.so. 4 分钟。

一个用于你的保险库的安全 REST API 和 Model Context Protocol (MCP) 服务器。

测试发现：npm install and npm run build succeed. All 1015 unit tests pass after fixing certificate key size; 1 test remains blocked on Node 18 ESM limitation. 17 分钟。

多引擎 MCP 服务器、CLI 和本地守护进程，用于 agent 网页搜索和内容检索，支持技能引导的工作流，无需 API 密钥。

测试发现：npm install and npm run build compile cleanly; 35/38 tests pass (3 live tests excused for external network limits); the daemon starts, health endpoint returns HTTP 200, and DuckDuckGo search returns real results; Playwright-dependent features are unavailable because the container runs Node 18 (requires 20+). 16 分钟。

一个中心，用于查找 Claude Skills、Agents、命令、钩子、插件和 Marketplace 集合，以扩展 Claude Code、Claude Desktop、Agent SDK 和 OpenClaw。

测试发现：npm install, validation of 294 agents/249 commands/40 hooks/89 skills, and 22 unit tests all pass after patching the test runner for Node 18 compatibility; the web-ui Next.js build and generate-registry scripts are blocked by missing Node 20+ and 3 absent script modules. 11 分钟。

使用 AI 控制 Gmail、Google Calendar、Docs、Sheets、Slides、Chat、Forms、Tasks、Search 和 Drive - 全面的 Google Workspace MCP 服务器和 CLI 工具

测试发现：workspace-mcp 1.29.0 installs, builds, launches in both stdio and streamable-http modes, serves a working MCP endpoint with 43 registered tools, passes 2660+ tests with mock Google API credentials, and both CLI entry points (workspace-mcp and workspace-cli) respond to --help. 31 分钟。

开源 AI agent 上下文层。PipesHub 将公司的知识（Slack、Drive、Jira、GitHub、Microsoft 365 及 50 多个连接器）转化为支持权限感知的工作区，agent 可以搜索、grep、导航和引用。支持 MCP、SDK 和内置 agent。可自托管。

测试发现：286 Python unit tests pass, 9633 JavaScript unit tests pass (4 pending). Both suites execute against mocked external services without real infrastructure dependencies. 62 分钟。

一个 Model Context Protocol （MCP）服务器和 CLI，为在 iOS 和 macOS 项目上工作的 agent 提供工具。

测试发现：XcodeBuildMCP v2.7.0 installs, builds (npm run build), passes all 2591 unit/integration tests across 243 test files, passes typecheck, and responds to CLI commands --help and tools on Node 18.19.1. 14 分钟。

用于 LinkedIn 的开源 MCP 服务器。让 Claude 和任何兼容 MCP 的 AI agent 访问个人资料、公司、职位和消息。

测试发现：Project installs cleanly with uv sync, server imports OK, test suite passes all non-environment-dependent tests (2921 of 3014 selected pass with the fix applied; the 11 failures are pre-existing container/TERM mismatches). 30 分钟。

Context7 Platform：面向 AI agent 的最新文档

测试发现：Build succeeds on all 6 packages. Test suites pass for CLI (361), MCP (59), SDK (27 with mock), and PI (4). The tools-ai-sdk has 11 of 16 tests passing; 5 require real AWS Bedrock credentials. The ctx7 CLI binary and MCP server both launch and respond correctly. 10 分钟。

开源 RAG 平台：内置引用、深度研究、支持 22 种以上文件格式、分区、MCP server 等。

测试发现：Agentset monorepo installs dependencies with bun 1.4.2 (using Bun's bundled Node 26.3.0), generates Prisma client with all enums, and passes 616 of 643 tests across web and engine packages. 42 分钟。

Agent OS：agent 会自行变得更聪明。我们只需守住底线：面试门槛、分阶段评估、预算化进化循环。MCP server，14 个运行时：Claude Code、Codex CLI、Gemini CLI、OpenCode、Copilot、Kiro 等。

测试发现：All unit tests pass (0 failures in ~21000 tests across 34 unit directories), ruff lint and format clean, CLI boots with correct version, MCP server help displays correctly, 2 Windows-specific bugs were caught and fixed. 73 分钟。

TalkToFigma：AI Agent（Cursor、Claude Code、Codex）与 Figma 之间的 MCP 集成，使 Agentic AI 能够与 Figma 通信，以读取设计并以编程方式修改设计。

测试发现：Dependencies installed, the project builds with tsup, the WebSocket relay starts on port 3055, and the MCP server starts on stdio, initializes, connects to the relay, and responds to all MCP protocol messages (initialize, tools/list, prompts/list, tools/call). 8 分钟。

Chrome MCP Server 是一个基于 Chrome 扩展的 Model Context Protocol (MCP) 服务器，它将你的 Chrome 浏览器功能暴露给 Claude 等 AI 助手，支持复杂的浏览器自动化、内容分析和语义搜索。

测试发现：The monorepo builds successfully (packages/shared, app/chrome-extension, app/native-server), all 642 tests pass, the CLI responds to help/doctor commands, and the MCP stdio server starts without errors. 12 分钟。

缺失的开源 Kubernetes UI，内置 MCP server，支持 AI agent。查看出了什么问题、原因以及变更。Issues、拓扑、事件时间线、Helm、GitOps、实时服务流量和集群审计，全部集成在一个 Go binary 中。

测试发现：Radar builds from source and runs; frontend is built and embedded, Go binary serves the UI at localhost:9282, and all unit tests pass. 21 分钟。

非官方且出色的 Home Assistant MCP 服务器

测试发现：Dependencies installed, smoke test passes, and the full unit test suite passes (11,790 passed, 357 skipped, 0 failed). 70 分钟。

使用任意 LLM（大型语言模型）进行深度研究。支持 SSE API 和 MCP 服务器。

测试发现：pnpm dependencies installed, next build completed with all 25 route handlers compiled, and the production server serves the Deep Research Next.js app at localhost:3000 responding HTTP 200 with a fully rendered page. 17 分钟。

官方 agent 插件，提供适用于 Azure 场景的技能和 MCP 服务器配置。

测试发现：Landing page builds with 0 errors and serves HTTP 200; all JSON metadata files validate; all 56 SKILL.md files are present; npm dependencies install cleanly with node 22. 4 分钟。

免费的 SEO MCP 服务器 + 开源的 SEO 和 GEO 技能（供 Claude 使用）：关键词研究、排名追踪、审计、反向链接、AI visibility 在你的真实 GSC/GA4/ads 数据上。

测试发现：All 8 SKILL.md files installed to ~/.claude/skills/ with correct content. The repository is a collection of Markdown skill definitions for Claude Code that rely on the external Ryze MCP connector, no local application to launch or tests to execute. 6 分钟。

透明高效的金融分析

测试发现：FinanceToolkit 2.2.0 is installed in a Python 3.12 venv, all 1472 tests pass, and both MCP CLI entry points (financetoolkit-mcp, financetoolkit-mcp-setup) respond correctly. 27 分钟。

HttpRunner 是一款开源的 API/UI 测试框架，简单易用，功能强大，具有丰富的插件化机制和高度的可扩展能力。

测试发现：无法验证；log 显示了它停在哪里。 86 分钟。

面向 Postgres、MySQL、SQL Server、Oracle、MariaDB、SQLite 的 Token 感知数据库 MCP 服务器。

测试发现：无法验证；log 显示了它停在哪里。 12 分钟。

测试发现：无法验证；log 显示了它停在哪里。

Draw.io Model Context Protocol (MCP) 服务器

测试发现：无法验证；log 显示了它停在哪里。

面向 AI 原生公司的自托管 AI Agents 平台。Claude Code、Codex、Gemini、OpenCode agents。Apache 2.0。

测试发现：无法验证；log 显示了它停在哪里。 29 分钟。

面向 AI Agent 和 MCP 服务器的下一代智能体式代理

测试发现：无法验证；log 显示了它停在哪里。 87 分钟。

MCP Gateway：一个轻量级网关服务，可立即将现有的 MCP 服务器和 API 转换为 MCP 服务器，无需更改代码。支持 Docker 部署和管理 UI，无需修改基础设施。

测试发现：无法验证；log 显示了它停在哪里。 87 分钟。