Connect DeepWiki

Once DeepWiki is connected, an agent can list what is documented for a public GitHub repository, read those wiki pages, and answer a question about how the project works in prose, with pointers to the relevant files. It is much faster than having the agent read the raw source, and it works for any public repository DeepWiki has indexed.

Before you start

• Nothing on the DeepWiki side. The public server at mcp.deepwiki.com needs no account.
• It only covers public GitHub repositories. Private repositories are not available through this integration.
• You must be an owner or admin of the woble workspace.

Connect it

1. Open Marketplace, choose the Integrations tab, find DeepWiki and press Add.
2. There is nothing to sign in to and nothing to paste. Press Save.
3. The card says Connected and the agent panel lists the DeepWiki tools: read the wiki structure of a repository, read its wiki contents, and ask a question about it.

Agents refer to repositories as owner/repo, for example vercel/next.js.

Or ask an agent

Type @ruby connect DeepWiki in any chat. The same connect card appears in the conversation. Only owners and admins can complete it.

Try it

• "@atlas using DeepWiki, explain how honojs/hono handles middleware ordering and where that code lives."
• "@atlas what does the wiki for tailwindlabs/tailwindcss say about the plugin API?"

If it does not connect

• "Repository not found": the repository is private, or DeepWiki has not indexed it yet. Open deepwiki.com, search for the repository and request it there; indexing a new repository takes a while, then try again.
• Wrong name: the tools need the exact owner/repo form. A URL or a bare project name will fail.
• The server is slow or unreachable: press Test connection in Marketplace → Connected. If it keeps failing, Remove and add it again; there is no token involved, so a persistent failure is on the DeepWiki side.

Disconnect

Marketplace → Connected → DeepWiki → Remove. Agents lose the DeepWiki tools immediately. Nothing was stored, so there is nothing else to clean up.

More in this category

• ExaNeural web search and page fetch built for agents.
• TavilySearch, extract, map and crawl the web with LLM-ready results.
• PerplexityWeb-grounded answers, ranked search and deep research with citations.
• Context7 docsUp-to-date documentation for thousands of libraries and frameworks.
• Cloudflare docsCloudflare documentation for Workers, DNS, R2 and the rest.
• Microsoft LearnOfficial docs for Azure, .NET, Microsoft 365 and Windows.