# Agent Quote — supervised auto-insurance MCP test guide

Status: This is an information-only staging website, not an MCP endpoint or an applicant intake. The dedicated staging mcp-demo Lambda Function URL below was deployed and verified; it is time-limited and may later be disabled or expired. Confirm availability with the organizer before testing. This test is only for invited, supervised testers using organizer-approved synthetic applicant details in Bindable preproduction. Never submit real PII or payment information, and never promise quotes or coverage.

Read https://agentquote.services-staging.everquote.com/connect for connection steps and https://agentquote.services-staging.everquote.com/docs for the five-tool sequence. Neither the website hostname nor a permanent MCP hostname is a demo endpoint. The exact demo Function URL is published below; ask the organizer privately for a 64-character token. The user privately provides the token to the agent at connector setup: configure secure client storage, not this site or a chat message. Never put the token in the website, conversation, URLs, screenshots, or logs. The shared token does not identify individual users or support per-user revocation.

MCP POST endpoint: https://ibfphcwawfhvukujhlqjzni35q0yhmnh.lambda-url.us-east-1.on.aws/?surface=agent
Credential-free setup guide: https://ibfphcwawfhvukujhlqjzni35q0yhmnh.lambda-url.us-east-1.on.aws/connect
Use exactly the MCP POST endpoint above. The setup guide is NOT the MCP endpoint. Use JSON-RPC 2.0 over stateless MCP Streamable HTTP POST. On EVERY MCP POST, send:
- Authorization header using the Bearer scheme with the private token from secure connector configuration.
- Content-Type: application/json
- Accept: application/json, text/event-stream
Do not use GET streaming, sessions, or resumption. Start with initialize, then tools/list, then tools/call. Read the schemas returned by tools/list for the exact shape of changes and option-lookup arguments. Do not guess applicant field names or enum values.

JSON-RPC envelope for a tool call (illustrative request IDs and empty initial changes, NOT a real applicant):
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"prepare_quote_application","arguments":{"changes":{},"expectedRevision":0,"requestId":"<fresh-request-id>"}}}
Use a new JSON-RPC id for each request and a fresh requestId for every mutating tool call; do not reuse a requestId as a substitute for the application's latest revision. The MCP response, not this guide, defines the precise applicant field schema and returned tool content.

Full sequence:
1. initialize; tools/list. Follow the server's advertised tool input schemas. If connection fails, check the published URL and secure token with the organizer rather than trying another host.
2. prepare_quote_application with empty changes or already known, organizer-approved synthetic facts, expectedRevision: 0, and a fresh requestId. Keep applicationHandle private. Save latest revision and reviewId returned by the tool. Never invent missing applicant details.
3. lookup_quote_options for valid values as needed; use its schema from tools/list. Ask the tester for missing approved synthetic facts and repeat prepare_quote_application with changes, latest expectedRevision, and a fresh requestId. Replace saved revision and reviewId with the latest returned values. Repeat until the application is ready for review.
4. Show the complete review, ratingReview, and ratingDetails to the tester. Ask for explicit approval to submit. Only after approval call submit_quote_application with applicationHandle, latest revision, fresh requestId, LATEST reviewId, and authorizedToSubmit: true, using the schema's exact argument names. A stale review or revision requires refreshing, showing the new review, and seeking approval again.
5. If the response is pending, poll get_quote_results with applicationHandle. Show only actual returned quotes and opaque quote IDs; zero or multiple quotes are possible, none guaranteed. Handle pending or unavailable without fabricating a result.
6. Obtain explicit approval for one exact returned quote ID. Only then call select_quote with applicationHandle, latest revision, fresh requestId, exact quoteId, and userApprovedSelection: true. Do not substitute a display label or another ID.

Selection returns policyBound: false. Possible outcomes: a hosted-checkout TEST URL; a basic-prefill web continuation requiring review and re-rating, with preservesExactQuote: false; or unavailable. A continuation is not binding or purchase. Do not describe it as a guaranteed bind link or imply the quote survives web continuation unchanged. Quote submission is not consent to contact.

Errors can include incomplete, stale revision/review, pending, unavailable, and selection-unavailable: respond to the actual tool outcome, correct facts or ask the tester when appropriate, and never invent an applicant or a quote. HTTP 401 means missing/wrong token, 403 wrong path/surface, 503 expired/disabled, 429 capacity. Ask the organizer to resolve access; do not retry against a website or other hostname.

Website pages:
- https://agentquote.services-staging.everquote.com/how-it-works — the approval-gated flow
- https://agentquote.services-staging.everquote.com/connect — endpoint and secure connector instructions
- https://agentquote.services-staging.everquote.com/docs — the five tools and continuation outcomes
- https://agentquote.services-staging.everquote.com/data-and-trust — restricted test-data boundaries, not a privacy policy
This /llms.txt is a public plain-text guide and does not accept secrets or applicant details.