signoz-searching-docs

SigNoz Docs

Use official signoz.io documentation and API references only. Ground every answer in fetched docs content and cite the canonical docs URL.

Access Docs

Prefer the SigNoz MCP server tools when available; fall back to direct HTTP fetch.

Preferred: MCP tools

• signoz_search_docs: keyword search over the indexed docs corpus. Pass 2 to 6 keywords for one topic as searchText, for example Kubernetes pod logs; keep product and technology names as the user wrote them and drop filler such as "how do I" or "please". Do not pass the whole question sentence. When the question contains only one usable term, such as Retention?, send that term alone rather than padding or inventing keywords. Split a question with two intents into two calls. Narrow with section_slug when the question maps cleanly to a single docs section; reuse a section_slug from an earlier result rather than guessing one, and defer to the tool's parameter descriptions for slug and ranking behavior. If the top result is off-topic, retry once with fewer or different terms before answering; for a single-term question, retry with a synonym or one added contextual term, such as data retention for Retention?.
• signoz_fetch_doc: markdown for one indexed page. Pass the canonical URL or /docs/... path; optionally narrow to a section with heading. Inspect truncation_reason and available_headings in every response rather than assuming the returned content is complete.

Never call signoz_search_docs with empty searchText; pass keywords taken from the user's question, adding a synonym or contextual term only for the retry described above. Never call signoz_fetch_doc without a URL; neither tool guesses missing input. signoz://... URIs are MCP resources: read them through the MCP resource API, never signoz_fetch_doc, which accepts only https://signoz.io/docs/... URLs or /docs/... paths and rejects other scopes.

Never construct /docs/... URLs from memory. Only pass URLs returned by signoz_search_docs to signoz_fetch_doc; if you think you already know the page path, search first and fetch the canonical URL from the result.

Fallback: direct HTTP fetch

If the MCP tools are unavailable, SigNoz docs support Accept: text/markdown natively.

Discover via the sitemap:

GET https://signoz.io/docs/sitemap.md

Fetch a specific page:

GET https://signoz.io/docs/<path>/
Accept: text/markdown

Workflow

1. Identify the domain from the user's question: instrumentation, OpenTelemetry setup, querying, dashboards, alerts, troubleshooting, deployment, or API docs.
2. Check the heuristics table below. If a heuristic matches, read it before answering: heuristics encode product decisions (which path/method fits the user's environment), useful in both paths.
3. Search and fetch: pick the path based on tool availability:
   • With MCP tools: call signoz_search_docs with 2 to 6 keywords from the user's question (or the single term when that is all it contains); pass section_slug if the domain maps cleanly to one. Read the top 1-3 results and call signoz_fetch_doc on the chosen URL (use heading to narrow if the page is large and the question is sub-section-specific). If the top result is off-topic, retry once with fewer or different terms (or a synonym for a single-term question).
   • Without MCP tools: grep sitemap.md for candidate pages, rank the best 2-5 by how directly they answer the task, and GET only URLs discovered in the sitemap with Accept: text/markdown. Heuristic coverage is sparse; for topics without a heuristic row, skim the sitemap by section path and prefer setup/troubleshooting/API-reference pages over overviews.
   • Fetch one page for narrow questions; fetch multiple pages when the task spans setup + troubleshooting, or method-selection + language guide. Keep the set small.
4. Handle truncated fetches before answering. When signoz_fetch_doc returns truncation_reason: "size", treat content as an incomplete prefix. Select the most relevant entry from available_headings and refetch the same search-result URL with that heading. If no heading covers the question, or the narrowed response is still truncated before the needed material, disclose that the fetched documentation is incomplete and do not infer that omitted content or a setting does not exist.
5. Answer from the fetched docs and cite canonical https://signoz.io/docs/... URLs.
6. Handle ambiguity deliberately: if multiple pages are plausible, prefer the one that completes the task most directly; mention alternates only when they materially change the answer.

Message Actions

On the terminal answer, emit FE-handoff actions per the SigNoz Skills & MCP spec:

• open_docs: include with the canonical URL of the primary cited page. Docs lookups are precisely the case where deep-linking to the source page helps the user read in context and verify the answer.
• follow_up: 1-2 next-step prompts that build on a docs answer. After a setup guide: "walk me through the first command" or "what's a common gotcha here?". After a concept page: "show me a worked example."
• Do NOT emit apply_filter. Docs answers do not produce a query for an explorer page; emitting apply_filter would overwrite the user's working query.

Verbatim guardrail: When answering a SigNoz docs question, include an open_docs action on the final message with the canonical URL of the primary cited page.

Domain Heuristics

Read the matching heuristic file before fetching docs. Each file contains decision logic to route the user to the right guide.