lockdocs
Pricing

lockdocsThe docs for the version you actually installed.

Exact-version library docs from your lockfile — local, offline, no rate limits.

Get started
Benchmarks
Pro pricing
GitHub
npx -y @sylphx/lockdocs setup
One question, two installs: pydantic 1.10.18 answers BaseModel.dict() (pydantic/main.py:427), pydantic 2.9.2 answers model_dump() (pydantic/main.py:352). It opens with Next.js 14 vs 15 cookies().

Exact version, zero config

Reads package-lock, pnpm, yarn, bun, Cargo.lock, uv, poetry, Pipfile, requirements and go.mod. No library IDs, no version hints in the prompt.

Docs that ship with the code

READMEs, changelogs and docs folders, plus API reference from .d.ts + JSDoc, Python docstrings, rustdoc and Go doc comments. Private packages included.

Upstream docs on first use

Next.js, Django or FastAPI ship no docs in the package? The first query adds their public GitHub docs at the commit your pinned release tag resolves to, anonymously and once. --no-fetch turns it off.

Offline after the first fetch

Package files come from node_modules, your virtualenv, the Cargo registry and the Go module cache. Once the model and first-use docs are cached it runs offline, and --offline blocks every download. No account, no quota.

Small, cited answers, fast

Hybrid search (BM25 + a small local embedding model) packs each answer into a 1,200-token budget, every section cited as package@version path:line. Median 58 ms per question after lockdocs fetch.

Three obvious tools

resolve (which versions), docs (a question, optionally scoped to a package), api (the exact signature of z.object or tokio::spawn).

Comparing documentation servers? ​

Lockdocs vs Context7 explains installed-version selection, offline queries and query quotas, with a reproducible two-version example. Copy-paste MCP setup covers Claude Code, Codex, Cursor, VS Code, Claude Desktop, Windsurf and Gemini CLI.