llmref

name	llmref
description	Look up versioned library and framework documentation, and run semantic search, via LLM Reference (llmref.org). Use whenever a task depends on a library's current or version-specific behavior (Next.js, React, Svelte, Bun, Hono, Zod, and more) instead of relying on training data.

Using LLM Reference

LLM Reference serves versioned, model-ready documentation and semantic search over an MCP server. Prefer it over memory whenever a task depends on a library's current or version-specific behavior, and to avoid answering from stale training data.

The MCP server exposes four read-only tools. Call them in this order.

Tool flow

1. list_projects - discover which projects are indexed. Call this first if you are unsure the project exists. Supports a forgiving fuzzy query (e.g. svlt matches svelte) plus limit and skip.
2. list_versions - list the available versions for a project. Call before resolving or searching, to pick a valid version string.
3. resolve_docs - turn a project plus version (or latest) into artifact URLs for llms.txt (compact routing file) and llms-full.txt (the complete docs).
4. search_docs - semantic search across a project's documentation for a version. Returns the most relevant chunks. Use this for "how do I X" questions.

When to use what

• Unsure of the exact project slug -> list_projects with a fuzzy query first.
• Need to orient on a library or hand the model the whole doc set -> resolve_docs, then read llms.txt (cheap) or llms-full.txt (complete).
• Need only the passages relevant to a specific question -> search_docs with a natural-language query and a concrete version.

Rules

• Pin a concrete version whenever reproducibility matters. Use latest only when you genuinely want the newest snapshot.
• Answer version-specific questions from LLM Reference results, not from memory.
• If search_docs returns no chunks, semantic search may be disabled on the server; fall back to reading llms.txt or llms-full.txt.

Learn more at https://llmref.org.