Skip to content
Answerworthy AI

MCP (external access)

The tool exposes a read-only MCP server at https://mcp.answerworthy.ai/mcp, so an AI assistant that speaks the Model Context Protocol (Claude Code, Claude Desktop, and similar) can pull your real numbers into a conversation without opening the app. There is no screenshot on this page because there is no screen — it is an API surface.

Two properties define it:

  • It can never change anything. Write tools are deliberately absent; every tool is a pure read of the same data the UI shows.
  • It requires a bearer token. Requests without one are rejected with a 401. Ask the platform operator for a token — this page deliberately says nothing more about where tokens live.

Connecting

Section titled “Connecting”

Point your MCP-capable assistant at https://mcp.answerworthy.ai/mcp (streamable HTTP) with the token as bearer authentication. Once connected, the tools below appear in the assistant’s tool list. Every tool that takes a brand argument accepts either the brand’s name (case-insensitive, partial match) or its UUID.

The tools

Section titled “The tools”
Tool What it returns
list_brands The tracked brands, with domains, industry, and how many questions each panel asks.
get_kpis The four executive KPIs — citation share (overall and per engine), crawler health, AEO readiness, AI-influenced revenue — as of the latest snapshot or a date you pass.
query_citations The URLs cited in the latest complete check, filterable by engine or domain substring.
get_prompt_results Per-question, per-engine status (cited / mentioned / absent, decided by the majority of replicates), filterable by question text.
get_share_of_voice Share of voice per engine for you and every tracked competitor, with the denominator spelled out.
get_crawler_stats Verified AI-crawler visits by day, lab, and class; spoofed visits are excluded.
get_gap_analysis The per-URL crawl-vs-citation join, each URL classified as gap, healthy, parametric, or dormant.
get_audit The latest on-page audit results with failing items and fix recipes; gate-blocked pages report the gate instead of a partial score.
get_fix_list The unified fix queue across all layers, sorted by impact and effort, filterable by status.
get_alerts The alert inbox, including whether the trend window is long enough for drift detection to be active.
search_answers Not implemented yet — semantic search over stored answers is a later phase, and the tool says so instead of pretending.

The same honesty as the UI

Section titled “The same honesty as the UI”

Every response carries the same evidence discipline the app uses, so an assistant reading these numbers inherits it:

  • An evidence label — observed (a reproducible measurement with a source reference) vs inferred (a professional read of the evidence). See OBSERVED vs INFERRED.
  • A provenance note — engine numbers come from API surfaces, a proxy for the consumer AI products rather than ground truth.
  • Honest nulls — anything not yet measured comes back as null with a reason. For example, get_kpis for a brand without a Cloudflare connection returns crawler health as null with a “not instrumented — unmeasured, not zero” note, and AI-influenced revenue is null until its data source lands, never modeled.
  • A trend-window caveat — until eight weeks of data accumulate, responses say the values are current levels, not trends.
Section titled “Related”