Ranked roundup · Documentation
Best MCP Server for Documentation
There is no single official MCP server for documentation — the category splits by whose docs you need. Microsoft is the only major docs publisher shipping a first-party server (a hosted endpoint at learn.microsoft.com/api/mcp, no account required), and it covers Microsoft docs alone. Everything general-purpose comes from startups and the community: Context7, run by Upstash, is the de facto standard at 61,000-plus GitHub stars; GitMCP turns any public GitHub repo into a free hosted docs endpoint; Ref sells token-efficient docs search with private-repo support; and LangChain's mcpdoc reads whatever llms.txt files you hand it. We verified every repository on September 1, 2026 — star counts, licenses, last-push dates, and auth models come from the GitHub API and vendor docs, not vendor claims. Rankings follow those verifiable criteria: official status, maintenance recency, adoption, auth friction, and coverage. No paid placement, and no invented benchmark scores.
Facts checked September 1, 2026 · ranked on verifiable criteria · placement is never for sale
Founder, Automation Squad ·
The ranking
- Official · Upstash — official for the Context7 service itself; the libraries it indexes are community-contributed
The default choice: current, version-specific library docs in any coding agent
resolve-library-id and query-docs tools; current, version-specific documentation and code examples for thousands of libraries, injected at query time
Hosted remote endpoint (https://mcp.context7.com/mcp) or local Node process via npx · No key required at low rate limits; free API key (Bearer header) raises them · MIT · 61,480★
- Docs corpus is community-contributed — the README itself says accuracy, completeness, and security of individual library docs can't be guaranteed
- Only the MCP server is open source; the API backend, parsing engine, and crawler are closed
- Rate-limited without an API key (the key is free)
- Official · Microsoft
Teams building on Azure, .NET, or Microsoft 365 who want first-party docs with zero setup
Search and fetch trusted, current Microsoft documentation and code samples; underlying index refreshes incrementally plus a full refresh daily
Fully hosted by Microsoft at https://learn.microsoft.com/api/mcp; nothing to install · None — public Streamable HTTP endpoint, no account or key, free of charge · CC-BY-4.0 · 1,867★
- Covers Microsoft documentation only — not usable as a general library-docs server
- Microsoft says to access it through an MCP client/agent framework, not as a traditional REST API
- Community
Pointing an agent at one specific GitHub project's docs with a zero-setup URL
Turns any public GitHub repository or GitHub Pages site into a docs endpoint, serving its llms.txt, readme, and docs to the agent
Free hosted service at gitmcp.io/{owner}/{repo}; self-hostable under Apache-2.0 · None — no signup, no key; the project states it doesn't store queries · Apache-2.0 · 8,363★
- Unofficial — no affiliation with GitHub despite the name
- Last push was May 8, 2026, roughly four months before our check
- Answer quality depends entirely on the target repo having decent llms.txt/readme/docs
- 4Official · Ref (ref.tools) — first-party server for a commercial product
Ref
Teams that need private repos and PDFs searchable alongside public docs, and will pay for it
ref_search_documentation and ref_read_url; token-efficient search over public docs plus private repos and PDFs, trimming fetched pages to roughly 5k tokens
Hosted HTTP endpoint (api.ref.tools/mcp) or local stdio via npx · API key required from the first request — sign-up needed even for the free trial · MIT · 1,164★
- Paid product: the free tier is 200 one-time credits, then plans start at $50/month (Pro) per the current pricing page
- API key required even to trial it
- Smallest adoption footprint on this list at roughly 1.2k stars
- Official · LangChain — their own open-source tool; serves any llms.txt sources you configure
Full control over sources: you decide exactly which llms.txt files the agent can read
Exposes a fetch_docs tool over a user-defined list of llms.txt files, so the agent reads only the documentation sources you choose
Local only (Python/npx); no hosted endpoint · None — runs on your machine; a domain allowlist restricts fetches to the llms.txt domains you list · MIT · 1,032★
- Only useful for sites that publish llms.txt, and you curate the list yourself
- Local-only — no hosted URL to paste into a client
- No token trimming — fetched pages land in context whole, unlike Context7 or Ref
Documentation MCP servers compared
| Server | Maintained by | Runs | Auth | License |
|---|---|---|---|---|
| Context7 | Upstash — official for the Context7 service itself; the libraries it indexes are community-contributed | Hosted remote endpoint (https://mcp.context7.com/mcp) or local Node process via npx | No key required at low rate limits; free API key (Bearer header) raises them | MIT |
| Microsoft Learn MCP Server | Microsoft | Fully hosted by Microsoft at https://learn.microsoft.com/api/mcp; nothing to install | None — public Streamable HTTP endpoint, no account or key, free of charge | CC-BY-4.0 |
| GitMCP | Community | Free hosted service at gitmcp.io/{owner}/{repo}; self-hostable under Apache-2.0 | None — no signup, no key; the project states it doesn't store queries | Apache-2.0 |
| Ref | Ref (ref.tools) — first-party server for a commercial product | Hosted HTTP endpoint (api.ref.tools/mcp) or local stdio via npx | API key required from the first request — sign-up needed even for the free trial | MIT |
| mcpdoc (LangChain) | LangChain — their own open-source tool; serves any llms.txt sources you configure | Local only (Python/npx); no hosted endpoint | None — runs on your machine; a domain allowlist restricts fetches to the llms.txt domains you list | MIT |
How to choose
Pick Context7 if you want the default: broad library coverage, a hosted endpoint, and no key required to start. Pick Microsoft Learn if your stack is Azure, .NET, or Microsoft 365 — it is the only first-party publisher option here. Pick GitMCP if you care about one specific GitHub project and want a zero-setup URL. Pick Ref if you need private repos and PDFs in the search index and can justify a paid plan. Pick mcpdoc if you want strict control over sources — it reads only the llms.txt files you list, locally.
How this ranking works
Every server here is real and was checked against its repository or vendor documentation on the date above — stars and licenses come from the GitHub API, not from memory. Order reflects verifiable criteria only: official support, maintenance recency, adoption, auth model, and documented capability coverage. We do not run paid placements, and we list every caveat we find, including the unflattering ones.
Common questions
- Is there an official MCP server for documentation?
- Not one that covers documentation in general. Microsoft Learn's hosted server is the only first-party option from a major docs publisher, and it covers Microsoft docs only. Context7 is officially run by Upstash, but the library docs it serves are community-contributed.
- What does setup and auth actually look like?
- Lighter than most MCP categories. Microsoft Learn and GitMCP are hosted URLs with no account or key; Context7 works unauthenticated at lower rate limits, with a free API key to raise them; only Ref requires sign-up and an API key from the first request, and mcpdoc runs locally with no account at all.
- Do any of these cost money?
- Ref is the only paid one: 200 one-time free credits, then plans from $50/month per its current pricing page. Context7, GitMCP, Microsoft Learn, and mcpdoc are free, though Context7 rate-limits unauthenticated use.
- Can my agent read private or internal docs?
- Ref is the only server here that advertises search over private repos and PDFs. mcpdoc can point at an internal llms.txt if your machine can reach it. Context7, GitMCP, and Microsoft Learn serve public documentation.
MCP servers we build
Automation Squad publishes its own open-source MCP servers and agent skills, documented with the same rules this ranking uses.