DeepWiki MCP server
The DeepWiki MCP server gives AI agents DeepWiki's generated documentation for public GitHub repositories, plus an ask tool that answers questions about a repo's code, free and without sign-in.
Written by
Connect DeepWiki
Claude Code
claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcpCursor and most other clients
{"mcpServers": {"deepwiki": {"url": "https://mcp.deepwiki.com/mcp"}}}Devin Desktop
{"mcpServers": {"deepwiki": {"serverUrl": "https://mcp.deepwiki.com/mcp"}}}No sign-in is needed. Clients differ on the URL field name: Devin Desktop uses serverUrl and most others use url, and a wrong field makes the client ignore the server silently.
Or paste this into your coding agent: Connect the DeepWiki MCP server (endpoint https://mcp.deepwiki.com/mcp) to my coding agent using the steps on https://docs.devin.ai/work-with-devin/deepwiki-mcp.
Your first question with this server installed
DeepWikiTool list 256One answer 5493 tools listed without a key
Typical serverFirst question 2,022 (median of 10 servers)
DeepWiki's first question takes 805 tokens: its tool list plus one answer. That is about 40% of the median first question of the 10 servers we measured with both numbers on (2,022 tokens).
Measured without a key or sign-in. Assistants that load every tool send this list to the model with each question; some load tools only when a task needs them. The one answer is from read_wiki_structure, a fixed read-only call. Tokens are counted with OpenAI's o200k tokenizer; other models count a little differently.
Measured . How we measure.
What the DeepWiki MCP server does
The DeepWiki MCP server connects an AI agent to DeepWiki, a site by Cognition, the company behind the Devin coding agent, that generates wiki-style documentation for public GitHub repositories. With it, an agent can list the documentation topics DeepWiki has written for a repo, read those pages, and ask a free-form question about the repo and get an answer grounded in its code, the same feature DeepWiki's site calls Ask Devin.
DeepWiki's server is useful when an agent has to work with an open-source project it does not know well: how a library structures its plugins, where a function is defined, or how two modules talk to each other. Because the answer comes from DeepWiki's index of the repository rather than from the model's memory, it reflects the code DeepWiki has indexed, not an older training snapshot.
Cognition runs the server at mcp.deepwiki.com with no API key and no account. The /mcp endpoint uses Streamable HTTP and is the recommended one; the older /sse endpoint is being retired. There is no public source repository for the server.
Tools it gives your agent
read_wiki_structureread_wiki_contentsask_question
When to use DeepWiki
- You want your agent to understand an unfamiliar open-source library before changing code that uses it.
- You need a quick answer to "how does X work in repo Y" without cloning the repo.
- You want a free docs tool that needs no key and works in any MCP client.
When to pick something else
- Private code: the free server only covers public repositories, and answers come from DeepWiki's index, which can lag behind the latest commits.
What the DeepWiki MCP server costs
Free hosted server with no sign-in, for public repositories only; private repositories need a paid Devin account and a separate server.
Current prices: devin.ai . We do not list prices because vendors change them often. Check the publisher's own page.
Questions people ask
Is the DeepWiki MCP server free?
Yes. Cognition's docs call it free and say it needs no authentication; it covers public GitHub repositories only (docs read ).
Does DeepWiki MCP work with ChatGPT or the OpenAI API?
Cognition's docs say the /mcp endpoint works with OpenAI and link OpenAI's remote MCP guide, which uses DeepWiki as an example for the OpenAI API. It also works in Claude, Cursor and Devin.
Can DeepWiki MCP read my private repo?
No. For private repositories Cognition points to the Devin MCP server, which needs a Devin account and API key.
Other MCP servers for docs and code context
- Context7: The Context7 MCP server gives AI coding agents current, version-specific documentation and code examples for thousands of libraries, fetched into the prompt so the agent stops guessing at outdated APIs.
- Microsoft Learn: The Microsoft Learn MCP server gives AI agents search and full-page reading over Microsoft's official documentation, plus official code samples for Azure, .NET and other Microsoft products, free and keyless.