AGENT DOCUMENTATION / SUPERVISED TEST
The staging MCP contract for invited, supervised testers. Tool schemas from tools/list remain authoritative for individual fields and changes.
The dedicated staging mcp-demo Lambda Function URL below was deployed and verified, but is time-limited. Confirm it is still enabled and obtain a private 64-character token from the organizer. Do not substitute a website hostname. See the connection guide and plain-text agent guide.
Start with empty changes or known synthetic facts, expectedRevision: 0, and a fresh requestId. Save the private applicationHandle, latest revision, and reviewId. Repeat with changes after lookup; use the latest revision each time.
Look up valid choices for incomplete or uncertain fields. Ask the tester for missing approved synthetic details; do not invent an applicant or guess option values. Prepare again to update the review.
Show the full review, ratingReview, and ratingDetails and obtain explicit approval. Submit with applicationHandle, latest revision, fresh requestId, latest reviewId, and authorizedToSubmit: true. Stale review or revision errors require a fresh review.
If submission is pending, poll with applicationHandle. Present only results actually returned, including opaque quote IDs. Pending, unavailable, or zero quotes are not guaranteed offers.
After showing quotes and obtaining explicit approval for one exact quote ID, send applicationHandle, latest revision, fresh requestId, quoteId, and userApprovedSelection: true. Selection returns policyBound: false, never proof of coverage.
Selection can return a hosted-checkout test URL, a basic-prefill web continuation that requires review and re-rating (preservesExactQuote: false), or an unavailable result. It never binds or purchases a policy. Handle incomplete, stale revision/review, pending, unavailable, and selection-unavailable responses without inventing data or promising quotes.
Read the test boundaries