LLM Reference

Versioned, model-ready documentation and semantic search for AI tools.

Give your AI accurate, version-correct docs instead of stale training data. Free, deterministic, and open - built on the llms.txt standard.

Docs · API · MCP registry · npm

---

Connect it to your AI tool

The quickest way to use LLM Reference is its MCP server - a remote endpoint that gives assistants four tools: list_projects, list_versions, resolve_docs, and search_docs.

https://api.llmref.org/v1/mcp

Client	Setup
Claude Code	claude mcp add --transport http llmref https://api.llmref.org/v1/mcp
Cursor	or add to .cursor/mcp.json
Codex	~/.codex/config.toml: [mcp_servers.llmref] then url = "https://api.llmref.org/v1/mcp"
Claude Desktop / Windsurf / VS Code	see the MCP guide

For any JSON-configured client:

{ "mcpServers": { "llmref": { "type": "http", "url": "https://api.llmref.org/v1/mcp" } } }

Full per-client setup lives at llmref.org/guides/mcp.

Or use the API, SDK, or CLI

One free public API at api.llmref.org - no account required.

# List indexed projects
curl https://api.llmref.org/v1/projects

# Fetch a library's docs, version-pinned and model-ready
curl https://api.llmref.org/v1/projects/drizzle/0.45.2/llms.txt

# Semantic search
curl -X POST https://api.llmref.org/v1/search/semantic \
  -H 'content-type: application/json' \
  -d '{"project":"drizzle","version":"0.45.2","query":"define a table schema"}'

Typed clients:

bun add @llmref/sdk       # TypeScript SDK
bun add -g @llmref/cli    # the `llmref` command-line tool

Every route, method, and type is documented at llmref.org.

What's indexed

Browse everything at /v1/projects or in registry/index.json.

Missing a library? Request it - you don't prepare anything; approved requests are ingested automatically into a versioned, searchable snapshot.

Request or fix documentation

Open an issue and the pipeline does the work - no setup on your end.

• 📥 Request docs - add a new project or version.
• 🐛 Report a problem - flag incorrect or stale content.

Approved requests become a versioned snapshot pull request that a maintainer reviews and merges before it goes live. Details: Contribute guide.

The registry (raw files)

Every artifact is also a plain, diffable file you can read straight from GitHub - no API required:

registry/<technology-id>/<version>/
  llms.txt        # compact routing/summary file for models
  llms-full.txt   # full merged documentation (only when the license permits it)
  meta.json       # snapshot metadata + content fingerprint + license
registry/index.json

llms-full.txt is published only when the upstream license permits redistributing full text; reserved-license snapshots ship llms.txt + meta.json and stay searchable via snippets. See SCHEMA.md.

https://raw.githubusercontent.com/reclear-io/llmref/main/registry/drizzle/0.45.2/llms.txt

Large dump artifacts (chunks, embeddings) live off-repo; see a snapshot's dumps endpoint. Format: SCHEMA.md · PR flow: CONTRIBUTING.md.

Links

• Documentation - https://llmref.org
• MCP registry - org.llmref/llmref
• npm - @llmref/sdk · @llmref/cli
• Code of Conduct · Security policy

License

MIT · Open Source by Reclear.io