THE AGENT JOURNEY / STAGING TEST
A supervised, synthetic-data test of the auto-only MCP flow. The demo was deployed but is time-limited; confirm it is still enabled.
Use the deployed, time-limited staging mcp-demo Function URL in the connection guide. Ask the organizer privately for a 64-character token. Configure the token securely in the MCP client, never in this website or conversation. Connect with ?surface=agent using stateless Streamable HTTP POST; call initialize, then tools/list.
Call prepare_quote_application with empty changes or known, organizer-approved synthetic facts, expectedRevision: 0, and a fresh requestId. Keep the returned applicationHandle private. Use lookup_quote_options for valid choices and repeat preparation with changes as needed, tracking the latest revision and reviewId. Never invent missing applicant details.
Show the tester the application review, ratingReview, and ratingDetails. Ask for explicit approval to submit. Only then call submit_quote_application with the private handle, latest revision, fresh requestId, latest reviewId, and authorizedToSubmit: true.
If the response is pending, poll get_quote_results with the applicationHandle. Show returned quotes and their opaque quote IDs; multiple quotes, or any quote at all, are not guaranteed. Do not turn an unavailable or incomplete response into a fabricated quote.
Obtain explicit approval for one exact quote ID before calling select_quote with the handle, latest revision, fresh requestId, quoteId, and userApprovedSelection: true. The result has policyBound: false: it may offer a hosted-checkout test URL, a basic-prefill web continuation requiring review and re-rating (preservesExactQuote: false), or no continuation. Selection is not purchase or binding.
Test only with invited, supervised participants and organizer-approved synthetic details in Bindable preproduction. Do not use real PII or payment data. The shared token is not a per-user identity and cannot be revoked for just one tester.
Read the data & trust notes