27 servers · data as of 2026-09-30
MCP server catalog
Ranked by GitHub stars, each integer read from the fetchable page recorded in that entry's source field, as of the asOf date. Star counts move daily, so every entry carries its own source and date.
stdio22
- Everything90,704Official MCP reference/test server: exercises every protocol feature — tools, resources and prompts — over both stdio and Streamable HTTP transports. The standard target for developing and testing MCP clients.2024-11-19MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Fetch90,704Official MCP reference server: fetches a given URL and converts web content to LLM-friendly Markdown; also ships a fetch prompt template.2024-11-22MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Filesystem90,704Official MCP reference server: secure file operations (read, write, create, move, delete, search) strictly inside the directories you explicitly allow, configured via command-line arguments or MCP Roots.2024-11-20MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Git90,704Official MCP reference server: lets the agent read and operate local Git repositories — status and diffs, commit history, branch creation and checkout — and includes repository-mutating tools such as git_add, git_commit and git_reset.2024-11-19MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Memory90,704Official MCP reference server: persistent cross-session memory as a knowledge graph (entities, relations, observations) stored in a local JSONL file (path set via MEMORY_FILE_PATH), exposed through both tools and a resource.2024-11-19MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Sequential Thinking90,704Official MCP reference server: a single sequential_thinking tool that structures problem solving into a dynamic, revisable thought sequence (branching, backtracking, revising earlier thoughts) — described officially as dynamic and reflective problem-solving.2024-11-27MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Time90,704Official MCP reference server: two tools — get_current_time for the current time in a given timezone and convert_time for timezone conversion. The smallest useful MCP server and a good teaching example.2024-11-26MIT / Apache-2.0（repo 正在授權轉移，見 repo LICENSE）
- Chrome DevTools MCP52,784Official MCP server from the Google Chrome DevTools team: lets a coding agent control and inspect a live Chrome — record performance traces with actionable insights, analyze network requests, read console messages (with source-mapped stacks), take screenshots, and automate reliably via Puppeteer.2025-09-11Apache-2.0
- Playwright MCP37,713Microsoft's official MCP server: drives a real browser with Playwright, understanding pages through the accessibility tree instead of screenshots. Tools click, type, navigate and snapshot; a persistent browser profile is the default, --isolated switches to in-memory sessions, and --port serves the SSE transport.2025-03-21Apache-2.0
- GitHub MCP Server33,297GitHub's official MCP server: connects AI tools directly to the GitHub platform — browse and query code, manage issues and PRs, analyze commits, monitor Actions workflows, review security and Dependabot alerts, and access discussions and notifications. Available as GitHub-hosted remote server (https://api.githubcopilot.com/mcp/, OAuth or PAT) or locally over stdio via the Docker image; tools are grouped into configurable toolsets.2025-03-04MIT
- AWS Documentation MCP Server9,744AWS Labs' official documentation server: read-only access to AWS documentation — fetch doc pages as Markdown, search docs via the official search API, read specific sections of large pages, filter and search rows in big doc tables (service quotas, pricing), and get content recommendations. Requires no AWS credentials at all.2025-04-03Apache-2.0
- Firecrawl MCP Server7,536Firecrawl's official MCP server: a full web-acquisition toolbox — firecrawl_search (query-first search), firecrawl_scrape (URL to clean Markdown/JSON), firecrawl_map (URL discovery), firecrawl_crawl (multi-page crawl with limit and path filters), firecrawl_interact (click/type page interactions), monitor tools (scheduled checks with diff alerts) and credit_usage (quota checks). A full profile registers 26 tools; there is also a keyless hosted endpoint (https://mcp.firecrawl.dev/v2/mcp, 3 tools) and a search-only endpoint (8 tools).2024-12-06MIT
- MCP Atlassian5,955Community-maintained (sooperset) MCP server for Atlassian products: brings Confluence and Jira to your agent, supporting both Cloud and Server/Data Center deployments; Cloud authenticates with email + API token, Server/DC with a Personal Access Token.2024-12-03MIT
- Obsidian MCP4,453Community-maintained (MarkusPfundstein) MCP server for Obsidian: operates your vault through the Local REST API community plugin — list files, read notes, full-text search, patch content at anchors, append content, and delete files.2024-11-29MIT
- Postgres MCP Pro3,352CrystalDBA's community Postgres MCP server: beyond schema browsing and SQL execution it ships DBA-grade tools — explain_query for execution plans (simulatable with hypothetical indexes), get_top_queries for slowest queries, analyze_workload_indexes/analyze_query_indexes for index recommendations, and analyze_db_health for comprehensive health checks. --access-mode switches between unrestricted (read/write) and restricted (read-only plus execution-time limits).2025-03-24MIT
- Tavily MCP Server2,416Tavily's official MCP server: four agent-oriented live-web tools — tavily-search, tavily-extract (structured extraction from pages), tavily-map (site structure maps) and tavily-crawl (systematic crawling). Run the local npx build or connect the remote https://mcp.tavily.com/mcp/ (OAuth flow supported).2025-01-27MIT
- MCP Server Kubernetes1,594Community-maintained (Flux159) Kubernetes MCP server: manage K8s clusters in natural language — list and deploy resources, apply manifests, manage pods and nodes, install/uninstall Helm charts. The docs explicitly list destructive tools (kubectl_delete, uninstall_helm_chart, cleanup_pods, node_management which can drain nodes, and kubectl_generic for arbitrary kubectl commands).2024-12-08MIT
- Docker MCP Gateway1,588Docker's official MCP toolkit and gateway (the docker-mcp CLI plugin): pick servers from the Docker MCP Catalog (hub.docker.com/mcp), each running in an isolated Docker container, and the gateway aggregates their tools, resources and prompts behind a single interface for all MCP clients — with built-in secrets management, OAuth flows, dynamic tool discovery and call tracing. Requires Docker Desktop 4.59+ with the MCP Toolkit enabled.2025-04-22MIT
- Qdrant MCP Server1,540Qdrant's official MCP server: turns the Qdrant vector database into a semantic memory layer for agents — qdrant-store saves information, qdrant-find retrieves by semantic similarity. Connects to a Qdrant server (QDRANT_URL) or a local embedded database (QDRANT_LOCAL_PATH); embeddings are computed locally with fastembed (default sentence-transformers/all-MiniLM-L6-v2).2024-12-02Apache-2.0
- Brave Search MCP Server1,478Brave's official MCP server: exposes the full Brave Search API surface to agents — web search, local business, places, images, video, news, LLM context and AI summarization. stdio is the default transport; BRAVE_MCP_TRANSPORT=http switches to HTTP (bound to 127.0.0.1:8080 by default), with allow/deny-list env vars for individual tools.2025-06-12MIT
- MongoDB MCP Server1,137MongoDB's official MCP server: operate MongoDB databases and Atlas clusters in natural language — query documents, inspect schema, perform database operations. A --readOnly flag restricts the tool surface to reads; Atlas management features can be wired with separate Atlas API credentials.2025-04-04Apache-2.0
- Redis MCP Server629Redis' official MCP server: manage and search Redis data in natural language — a full tool surface over strings, hashes, lists, sets, sorted sets, JSON documents, pub/sub messaging and Redis Streams, positioned as the data-layer interface for agentic applications.2025-04-02MIT
http5
- Context762,552Upstash's Context7: up-to-date, version-matched library documentation for LLMs and AI code editors. Two MCP tools — resolve-library-id maps a library name to a Context7 ID, and get-library-docs pulls that version's docs and code examples — reducing hallucinated, outdated APIs from model memory. Remote server at https://mcp.context7.com/mcp with an Authorization: Bearer API key header.2025-03-26MIT
- Exa MCP Server5,065Exa's official MCP: connects agents to Exa's neural search — default tools web_search_exa (search with clean, ready-to-use content) and web_fetch_exa (pages to Markdown), plus optional tool sets and Exa Agent. The hosted endpoint https://mcp.exa.ai/mcp works anonymously with rate limits; OAuth or an API key unlocks higher limits.2024-11-27MIT
- Stripe MCP1,849Stripe's official hosted remote MCP server (https://mcp.stripe.com): over an OAuth grant it gives MCP clients access to your Stripe billing infrastructure on your behalf, and the official docs show building autonomous agents with it. Code and SDKs live in the stripe/ai repository.date not on recordMIT（stripe/ai 倉庫）
- Sentry MCP871Sentry's official remote MCP server (https://mcp.sentry.dev), designed for human-in-the-loop coding agents with a focus on developer debugging workflows — query and analyze Sentry errors, issues and traces. The AI-powered search tools additionally need an LLM provider. A stdio transport also exists (officially a work in progress, mainly for self-hosted Sentry).2025-03-29FSL-1.1（Functional Source License，未來版本轉 Apache-2.0）
- Neon MCP Server649Neon's official remote MCP server (https://mcp.neon.tech/mcp): operate Lakebase Postgres on Neon in natural language — create projects and branches, manage endpoints, run queries, inspect schema and run database migrations. Source is open at neondatabase/mcp-server-neon.2024-11-29MIT