DeepWiki MCP

DeepWiki MCP is a remote Model Context Protocol server that lets an AI client read the documentation DeepWiki generates for public GitHub repositories and ask questions about those repositories. It is hosted at https://mcp.deepwiki.com/mcp, is free and needs no authentication, according to Devin's documentation. There is nothing to install: you add the URL to your client.

Quick facts (as of November 3, 2026)

Tools

The documentation lists three tools:

The documentation page does not list the input parameters for these tools, so check your client's tool list after connecting.

Setup

Use the /mcp endpoint. The SSE endpoint is described as legacy and being deprecated.

Claude Code

claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcp

Cursor and most other clients

Per the documentation, most clients such as Cursor accept only a url field:

{
  "mcpServers": {
    "deepwiki": {
      "url": "https://mcp.deepwiki.com/mcp"
    }
  }
}

Devin Desktop

Devin Desktop accepts either serverUrl or url:

{
  "mcpServers": {
    "deepwiki": {
      "serverUrl": "https://mcp.deepwiki.com/mcp"
    }
  }
}

The documentation warns that using a field name your client does not recognize causes the server to be silently ignored. If the tools do not appear, check whether your client wants url or serverUrl.

When it helps

Use it when you want an assistant to explain an unfamiliar open source project, find where a feature lives in a codebase, or answer "how does this library handle X" without cloning the repository. Because the answers are generated, treat them as a starting point and confirm important details in the repository itself.

Example prompts

Once connected, you can ask your assistant things like: "List the documentation topics DeepWiki has for this repository," "Read the DeepWiki page for how this project handles configuration," or "Ask DeepWiki how this library structures its plugin system." Tell the assistant which public GitHub repository you mean. The documentation page does not give input formats, so check the tool schema your client shows for the exact parameters.

DeepWiki MCP compared with other ways to give a model code context

Reading a repository through DeepWiki is different from letting a model read local files or search a Git host. DeepWiki returns generated documentation and answers for public projects, so it is quick for orientation but is a second-hand summary. A filesystem or Git server reads the actual code you point it at, which is slower to set up but exact. Many people use both: DeepWiki to find where to look, then the source to confirm.

Limitations and things to watch

• Public repositories only. DeepWiki MCP does not read private repositories.
• Generated answers. ask_question returns an AI-generated response. It is grounded in the repository, but it can still be wrong.
• SSE is on its way out. Use the /mcp endpoint.
• Silent failure on the wrong field name. A config with a field your client does not support is ignored without an error.
• Hosted service. The server runs at mcp.deepwiki.com, so requests go to that remote service rather than a process on your machine.

Pros and cons

Pros

• Free, no account, no API key.
• Nothing to install: one URL.
• Three focused tools for reading and questioning repository documentation.

Cons

• Public GitHub repositories only.
• Answers are AI-generated.
• Tool input parameters are not listed on the documentation page.
• Private repositories need a different product.

Who it is for

A good fit if:

• You want an assistant to explain a public repository you have not read yet.
• You compare open source libraries and want quick architecture summaries. For library documentation specifically, see Context7, which serves docs for libraries to the model.

Not the right tool if:

• Your code is in a private repository. Use your client's own code tools, or a server for your Git host such as the GitLab MCP server.
• You need to read local files. The Filesystem server is for that.

What we could not verify

• Tool parameters. Not documented on the Devin docs page.
• Rate limits or usage limits. The page we read does not mention any.
• Which repositories are indexed. The page says public GitHub repositories; it does not say how fresh the generated documentation is.
• How it behaves in practice. The page describes the service; this listing is based on that documentation alone.

How this page was put together

We read Devin's DeepWiki MCP documentation on November 3, 2026 and compared the endpoints, tools and client configs it lists.

Sources

• Devin docs: DeepWiki MCP