ErrLookup

Know what every error means — and how to fix it.

ErrLookup reads the source of popular open-source libraries and documents every user-facing error they can raise: the exact message, what triggers it, ordered fixes, and defensive patterns — pinned to the commit that was analyzed.

Live: errors.standardbeagle.com · Demo video · Blog: how it works

For coding agents (MCP)

{
  "mcpServers": {
    "errlookup": { "command": "npx", "args": ["-y", "@standardbeagle/errlookup-mcp"] }
  }
}

The MCP server downloads the versioned static dataset once, caches it locally, and answers search_error / get_error from disk — offline-capable, no API keys.

For everything else

• REST: GET https://errors.standardbeagle.com/api/search?q=<raw error message> — tiered matching (exact code → derived pattern → fuzzy), scored results with citations.
• Static dataset: CORS-enabled versioned JSON at /data/manifest.json — the entire contract, no server required. See API docs.
• Humans: one page per error with the exact message text, source permalink, and fixes.

Why source-derived beats forum search

Forum answers are point-in-time guesses about somebody else's version. Every ErrLookup record is mechanically tied to a file, line, and commit SHA — regenerated when the code changes, honestly labeled as AI-assisted analysis, with the exact source line one click away. This matters most for compiled languages: when a Go binary or a JAR throws in production, the error string in your logs is the only artifact you have — the source never shipped.

How it works

repo @ SHA ──lci/regex extraction (0 tokens)──▶ candidate sites
   ──batched LLM classify/enrich/defend──▶ validated records
   ──independent verify pass──▶ patched records
   ──atomic export──▶ versioned static JSON ──▶ CDN ──▶ site / MCP / API

• Deterministic first: a structural grep (via lci or a built-in walker) finds error-raising sites across nine language families before any model runs. Models judge concrete sites; they don't wander repos.
• Model-agnostic: providers are pluggable subprocesses (CLI print mode or ACP); each phase routes independently — cheap models do bulk work, a stronger one verifies. Benchmark results.
• Crash-safe: phases checkpoint per repo+SHA with persisted outputs; exports are atomic; a torn dataset is impossible.

Self-hosting for internal repos

The pipeline is a local CLI and the dataset is static files — analyze private code on your own hardware and serve it from any internal host. Walkthrough: Run ErrLookup on your own repositories.

Workspace

package	what
packages/schema	zod schemas + generated JSON Schema — the single source of truth
packages/pipeline	analysis CLI: extraction, LLM phases, SQLite state, atomic export
packages/site	Astro static site + Pages Functions API
packages/mcp	@standardbeagle/errlookup-mcp — stdio MCP server

pnpm install
pnpm test                                  # all suites
pnpm --filter @errlookup/pipeline dev analyze sindresorhus/is
pnpm --filter @errlookup/pipeline dev export

Requesting a library

Request a crawl or open an issue with the crawl-request label.

---

Built by Standard Beagle Studio — the AI UX agency for bold product teams. MIT licensed.