Answer Radar From Claude, Cursor and Other MCP Clients

Connecting

Start at /mcp, which describes the connector and links the setup guides. The docs at /mcp/docs cover the quickstart, authentication, connecting Claude, connecting Cursor, the tool reference, rate limits and troubleshooting. Claude.ai connectors use the OAuth 2.1 authorization code flow with PKCE; the authorization endpoints are published at the standard well-known paths. Cursor and other clients follow the same remote-server pattern.

The Answer Radar tools require the Compete plan, the same entitlement the web pages check. A client without it gets an entitlement refusal, not silent empty data.

The Answer Radar tools

Projects: answer_projects_list, answer_project_get, answer_project_create, answer_project_update. Prompts: answer_prompts_list, answer_prompts_propose, answer_prompts_update. Audits and runs: answer_audit_estimate, answer_audit_start, answer_audit_launch, answer_runs_list, answer_run_get. Measurement: answer_metrics_get, answer_trends_get, answer_fanout_get, answer_gaps_list, answer_gap_get, answer_content_gaps_get, answer_sources_fetch. Fixes: answer_fix_draft, answer_fix_check, answer_fix_update, answer_fix_save, answer_fix_get, answer_fixes_list. Verification: answer_verify_start, answer_verify_get. Operations: answer_actions_list, answer_action_get, answer_action_update, answer_shares_list, answer_share_create, answer_share_revoke, answer_schedule_set, answer_usage_get, answer_activity_list, answer_digest_get. Search Console: gsc_queries_get.

Read tools carry the answer read scope and write tools the write scope. Most tools accept a mode: summary for the numbers, evidence for the recorded facts behind them. Every metric response carries its definitions and disclosure block, so your client can caveat a number without a dashboard from us.

A first workflow

Ask your client to list projects, get one project's configuration, and list its prompts. Estimate an audit, check usage, then start the audit and poll the run. When it completes, get metrics and read the disclosure with them, list gaps ranked by intent and evidence, and open one gap in full. Draft a fix skeleton for it, write the fix, check it against the evidence, update it under the guard, and save it to the Fixes workspace for a person to approve. After publishing, register a re-verification and read the delta a week later.

That loop is the product. The web app is where a human approves, and the client is where the work happens, which is why approve and publish exist only in the app.

• Use evidence mode when you need the cited pages, not just the counts.
• Pass an idempotency key on writes; retries then finish rather than duplicate.
• Read the disclosure before quoting any number.

FAQ