docs-lookup

Cognigy Docs Lookup

The plugin bundles the official Cognigy documentation MCP server (docs.cognigy.com). Use it as the source of truth for platform questions — the platform changes faster than model training data. Prefer documented answers over prior knowledge, and link the doc page you used.

Default to the filesystem tool, not search. Cognigy's docs pages are very large (parameter-details.mdx alone is 62 KB) and search_cognigy_documentation returns matched pages whole rather than as excerpts, so its payload scales with page size instead of relevance. Measured: a Voice Gateway barge-in query returned 90 KB via search vs ~4 KB via rg; a conceptual "which tool does the AI Agent call" query returned 280 KB via search — exceeding the tool-result limit and yielding nothing usable — vs ~3 KB via rg.

Tools

• query_docs_filesystem_cognigy_documentation — the primary tool. Read-only shell-like queries over a virtual filesystem of every docs page (.mdx) plus OpenAPI specs. The server documents support for rg, grep, find, tree, ls, cat, head, tail, stat, wc, sort, uniq, cut, sed, awk, jq and basic text utilities; run --help on any command to confirm usage. Flag coverage is narrower than GNU (e.g. cat -A is rejected), so keep flags conventional. These run inside the MCP tool's virtual filesystem — NEVER in the local shell. Each call is stateless: the working directory resets to /, so use absolute paths or chain with && (cd /api-reference && ls).
• search_cognigy_documentation — semantic search. Takes only a query string: no result limit, no relevance floor, no filters. Fall back to it only when you cannot guess the docs' vocabulary at all. Expect a very large result; when it overflows into a file, triage by extracting the Title:/Page: lines, then read the best one or two pages with the filesystem tool — never read the dump whole.
• submit_feedback — report incorrect, outdated, or confusing docs to the Cognigy docs team. Offer this when you and the user hit a genuine documentation gap.