Check that the server is up and report its version.

Fetch an MDN Web Docs or TypeScript handbook page as markdown.

identifier resolution order:

1. "mdn:" like "mdn:Web/JavaScript/Reference/Global_Objects/Array" → fetched directly from the MDN docs path;

2. "ts:" like "ts:intro" or "ts:2/generics" → fetched directly as a TypeScript handbook page;

3. plain names like "Promise" → resolved through the search index: only the top-score candidates are considered; a single distinct path wins, ties prefer mdn Web/JavaScript/Reference pages over mdn Web/API pages over other mdn pages over typescript pages (within the winning tier a parent page beats its subpages); ambiguous or missing matches return an error dict suggesting js_search().

topic (a heading like "Methods") keeps only that heading section plus title/description; if no such section exists the full content is returned with a "note" listing the available headings. max_tokens truncates the final markdown to roughly that many tokens (estimated as len(text)//4), cut at a line boundary, and sets "truncated": true when applied.

On failure returns {"ok": False, "error", "suggestion"} — try js_search() to find the exact page, or use the mdn:/ts: prefix.

Search MDN Web Docs and the TypeScript handbook by name/path.

Ranks entries of the cached search index (every query token must match; exact name > startswith > substring in name > substring in path). Returns {"ok": True, "query", "results": [{source, name, path, score}], "index_count", "stale"}. When the index is unavailable returns {"ok": False, "error", "suggestion"} — retry later.

Look up a package on the npm registry (metadata + readme).

Returns name, resolved version, description, license, homepage, repository URL, keywords, engines, dependencies and the readme as markdown. When version is given the lookup is pinned to that release; otherwise the latest version is used. Metadata is cached for 1 day keyed by name+version; failures are never cached.

npm no longer puts the README in the packument — for modern packages the readme field is an empty string — so the document is fetched from the package's GitHub repository. When it is found, readme_source names the exact file (e.g. github:colinhacks/zod@HEAD/packages/zod/README.md); when it is not, readme_markdown stays null and note says why. Only github.com repositories are followed; gitlab/bitbucket packages and repos with no repository URL get the note, not a guess.

max_tokens truncates the README to roughly that many tokens (estimated as len(text)//4), cut at a line boundary, and sets "truncated": true when applied. The cache keeps the full README, so the same entry can answer a later call with a larger max_tokens.

On failure returns {"ok": False, "error", "suggestion"}.

Real health check: search index, local cache and upstream endpoints.

The cache check gains read_only: true plus read_only_reason when the database cannot be written (P5). It stays an "ok" check on purpose: a cache that only reads is a degraded optimisation, not a sick server, so overall does not change.

Probes developer.mozilla.org, typescriptlang.org and registry.npmjs.org with a light GET (8s timeout each, through the politeness layer — no curl subprocess). Never raises; returns {"server", "version", "checks": {...}, "politeness": {...}, "overall": "ok"|"degraded"|"error"} where overall is "error" when the search index is unavailable, "degraded" when any check fails, and "ok" otherwise. The "politeness" block holds the layer's counters (robots cache state, throttle delays, budget use, conditional-GET hits) and is not a check: it never changes "overall".