Your AI's memory palace for docs that actually exist

What it does

Grounded Docs is a local documentation index that plugs into any MCP-compatible client (Claude, Cline, Copilot, etc.). It scrapes docs from websites, GitHub, npm, PyPI, or your filesystem, then answers your AI’s questions with the exact version you’re using. Think of it as a reference librarian that lives on your machine and never sleeps.

The interesting bit

The scraping is surprisingly thoughtful. It probes for llms.txt files, requests Markdown via content negotiation, and even auto-upgrades to Playwright when it hits hash-routed SPAs. The format support is almost comically broad — 90+ source languages plus Office documents, Jupyter notebooks, EPUBs, and Bazel configs. Someone really wanted to index their entire hard drive.

Key highlights

• Runs fully offline; code and queries never leave your network
• Optional semantic search via pluggable embeddings (OpenAI, Ollama, Gemini, Azure)
• CLI-first design with structured JSON/YAML output for agent consumption
• Includes “Agent Skills” — pre-written prompts that teach AI assistants how to use the tool
• Web UI on port 6280 for manual index management
• Benchmarking suite with IR metrics and LLM-judged relevance scores

Caveats

• Requires Node.js 22+ for the CLI path; Docker available but not the default experience
• Semantic search is optional but the README warns quality “dramatically” drops without it
• Hash-routed SPA handling requires explicit flags; auto-detection would be nice but isn’t claimed

Verdict

Worth a look if you’re tired of your AI confidently explaining APIs that shipped three years ago. Skip it if you already pay for Context7 or Nia and don’t mind the lock-in — this is the escape hatch, not the luxury suite.