mcp-deepwiki

What it does

This server crawls deepwiki.com documentation sites, converting HTML to Markdown for LLM consumption. It sanitizes content by stripping headers, footers, navigation, scripts, and ads, then returns either a single aggregate document or structured by-page data. Domain safety restricts processing to deepwiki.com URLs only. It supports configurable crawl depth, concurrent fetching, and request timeouts.

Who it's for

Engineers building documentation-aware AI tools, researchers needing structured library reference material for analysis, and developers embedding external knowledge into Claude prompts for context-aware reasoning and decision support.

Common use cases

• Fetch shadcn/ui or Tailwind CSS documentation as Markdown for model context
• Extract library architecture and API patterns for analytical tasks
• Build documentation-backed RAG and embedding systems
• Pre-process reference material for prompt engineering and knowledge grounding

Setup pitfalls

• Currently broken: DeepWiki has blocked scraping. The maintainers recommend using the official DeepWiki MCP server at docs.devin.ai instead.
• Filesystem access: Requires read/write permissions for caching and operation state. Run in a sandboxed environment if processing untrusted content.
• Network connectivity: Makes outbound HTTP requests to deepwiki.com. Firewalls, proxies, and rate-limiting may block requests. Configure DEEPWIKI_REQUEST_TIMEOUT (default 30s) and DEEPWIKI_MAX_RETRIES (default 3) to handle network instability.
• Maintenance and configuration: CI builds are failing and the last commit was 77 days ago. Tune DEEPWIKI_MAX_CONCURRENCY (default 5) and DEEPWIKI_RETRY_DELAY (default 250ms) for your infrastructure.