DeepWiki MCP Server
AI-powered search over GitHub repo documentation, exposed as a free remote MCP server.
Updated
What is DeepWiki MCP?
The DeepWiki MCP server is a Model Context Protocol server that gives AI agents searchable, AI-generated documentation for public GitHub repositories indexed on DeepWiki.com.
DeepWiki, built by Cognition (the team behind Devin), automatically turns public GitHub repos into structured, wiki-style knowledge bases by combining large language models with code analysis. The MCP server exposes that knowledge programmatically, so a coding agent can read a repo's documentation structure, pull its contents, or ask natural-language questions and get context-grounded answers.
The server is officially hosted by Cognition at https://mcp.deepwiki.com/mcp (Streamable HTTP). It requires no authentication — public repositories are accessible without a login or API key, which makes connecting an agent a one-line setup.
Tools the DeepWiki MCP server exposes
| Tool | What it does |
|---|---|
| read_wiki_structure | Get the list of documentation topics (table of contents) for a GitHub repository. |
| read_wiki_contents | View the generated documentation contents for a GitHub repository. |
| ask_question | Ask a natural-language question about a repository and get an AI-powered, context-grounded answer. |
Connect the DeepWiki MCP server
Claude Code
- 1
Add the remote server
Run: claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcp
- 2
No auth needed
The server is public and requires no login, API key, or OAuth.
- 3
Verify
Run /mcp in Claude Code to confirm the deepwiki server is connected, then ask about any indexed GitHub repo.
Cursor
- 1
Open MCP settings
Go to Settings > MCP (or edit ~/.cursor/mcp.json).
- 2
Add the server
Add: { "mcpServers": { "deepwiki": { "serverUrl": "https://mcp.deepwiki.com/mcp" } } }
- 3
Reload
Save and reload Cursor; the deepwiki tools appear with no authentication step.
Sample use cases
An agent needs to understand an unfamiliar open-source dependency before writing integration code.
It asks DeepWiki questions about the repo and gets context-grounded answers without reading the whole codebase.
A scheduled workflow tracks a fast-moving library.
The agent pulls the latest documentation structure and summarizes changes for the team.
A code review agent encounters an unfamiliar library import in a pull request
It calls ask_question about the library's API patterns and returns context-grounded review comments citing official documentation.
An onboarding agent answers a new developer's question about a large internal monorepo
It uses read_wiki_structure to find the relevant section, then read_wiki_contents to fetch the explanation — no codebase traversal needed.
A migration planning agent compares two framework versions
It asks DeepWiki about breaking changes between versions and generates a migration checklist grounded in the actual documentation.
An architecture agent maps how multiple microservices interact
It reads wiki structures for several repos, cross-references API surfaces, and produces a dependency diagram with documentation links.
A security review agent checks whether a dependency follows safe defaults
It queries DeepWiki about the library's authentication and input validation patterns and flags deviations from best practices.
Security & permissions
The DeepWiki MCP server uses no authentication — there is no login, API key, or OAuth flow for public repositories. The agent only receives read access to AI-generated documentation for repos that are already indexed on the public DeepWiki.com site; it does not get write access to GitHub or to any private code. Indexing a private repository is a separate, account-gated flow handled through Devin, not through the unauthenticated MCP endpoint.
Using DeepWiki MCP with Gamut
In a Gamut workflow, the DeepWiki MCP server is wired in as a read-only research tool that an agent can call on a schedule or in response to an event — for example, when a new dependency is added or a tracked repo ships a release, an agent can call read_wiki_structure and ask_question to summarize how a library works and feed that context into a downstream step. Gamut manages the connection to https://mcp.deepwiki.com/mcp (no credentials needed), so the agent just invokes the tools. Note that results are limited to public repositories already indexed on DeepWiki.
Frequently asked questions
What is the DeepWiki MCP server?
It is an official, remote Model Context Protocol server from Cognition (makers of Devin) that lets AI agents read AI-generated documentation and ask questions about public GitHub repositories indexed on DeepWiki.com. It exposes three tools: read_wiki_structure, read_wiki_contents, and ask_question.
Is the DeepWiki MCP server free?
Yes. The public server is completely free with no login or authentication required. Indexing private repositories requires a Devin account, and open-source maintainers can apply for Devin credits separately.
How do I connect DeepWiki MCP to Claude Code or Cursor?
For Claude Code, run: claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcp. For Cursor, add { "mcpServers": { "deepwiki": { "serverUrl": "https://mcp.deepwiki.com/mcp" } } } to your MCP config. No authentication step is needed.
Is DeepWiki MCP hosted or self-hosted?
It is officially hosted by Cognition as a remote server at https://mcp.deepwiki.com/mcp (Streamable HTTP). You connect to the hosted endpoint directly rather than running it yourself.
Can DeepWiki MCP access private GitHub repositories?
Public repositories are indexed automatically and freely accessible. Indexing private repositories requires a Devin account from Cognition. Open-source maintainers can apply for Devin credits separately.
How current is the documentation that DeepWiki indexes?
DeepWiki re-indexes repositories periodically. For popular public repos, documentation is generally kept up to date. You can check freshness by comparing the wiki structure against the latest repo commits.
What tools does the DeepWiki MCP server expose?
Three tools: read_wiki_structure (table of contents for a repo), read_wiki_contents (full documentation pages), and ask_question (natural-language Q&A grounded in the repo's documentation).