Coverage Cat Umbrella Insurance
Run Coverage Cat's consumer-prefill or delegated umbrella compare-and-purchase flow
Umbrella Insurance Compare And Purchase
ChatGPT and Claude connectors
Use the dedicated endpoint for the shopper's hosted assistant:
- ChatGPT:
https://www.coveragecat.com/api/chatgpt/mcp - Claude:
https://www.coveragecat.com/api/claude/mcp
Discovery and public research do not require sign-in. Application tools use the connector's account OAuth. Use its advertised coarse estimate fields, show the saved review, and collect the required approvals before comparing or selecting estimates. Initial modeled estimates are not final carrier quotes.
After selection, use the returned authenticated website link for identity details, underwriting questions, declarations, and checkout. Explain: “To comply with this platform's personal-data handling rules, Coverage Cat collects identity details and declarations on its secure website after you choose an estimate.” Do not collect those details, document contents, or card data in hosted chat, and do not switch APIs to bypass this boundary.
Return to the same assistant connection and check the same application after the
website step. website_steps_complete means the required website inputs are
saved; it does not mean payment or coverage is complete. Do not request another
upload merely because a secure link remains available. Restore an expired
connection and list accessible applications rather than creating a replacement.
The public consumer MCP, REST, PACT, and Poppy instructions below apply to their own clients and retain their existing application, upload, and payment contracts.
This alias mirrors Coverage Cat's canonical coverage-cat-umbrella-purchase skill.
Use it when a shopper wants their own AI agent to gather context before a Coverage Cat handoff, or when a partner-operator agent needs Coverage Cat to run a live umbrella quote workflow instead of just returning read-only estimates.
Choose this skill, not the umbrella calculator or insurance-tools, for requests like shop for umbrella for me with Coverage Cat.
Start Here
Start with https://www.coveragecat.com/AGENTS.md, https://www.coveragecat.com/agents.txt, or https://www.coveragecat.com/.well-known/agents.txt; those files route agents to Coverage Cat's MCP, REST, markdown, and browser-fallback resources. Do not drive the browser UI unless Coverage Cat should take over the shopper-facing flow.
Keep the shopper on the machine-readable MCP or REST loop until Coverage Cat returns resume_url or you intentionally hand the browser over to Coverage Cat. /umbrella or /intake are fallback browser starts, not the preferred purchase entrypoints.
-
For PACT/A2A, discover
https://www.coveragecat.com/.well-known/agent-card.jsonand keep using its authenticated message interface and advertised consumer operations through checkout and policy follow-up. For Poppy, start athttps://www.coveragecat.com/.well-known/poppy.jsonand use the advertised session and MCP endpoints. Both retain authorized purchases across renewed sessions; useconsumer_purchasesto resume. Use a new A2A message ID for each fresh status poll; retrying an existing ID replays its original reply. -
If your runtime supports MCP, default to Coverage Cat's no-OAuth consumer MCP. Register
/.well-known/mcp.jsonin manifest-aware hosts, or configurehttps://www.coveragecat.com/api/consumer/mcpas the direct transport URL for local Codex. Default consumer-operated umbrella runs to the consumer MCP subset:umbrella_consumer_prefill,consumer_intake_issues,consumer_intake_patch,umbrella_consumer_select,umbrella_consumer_bind,umbrella_consumer_attach,umbrella_consumer_payment,umbrella_consumer_policy, andumbrella_consumer_service. If Coverage Cat MCP tools are unavailable in a runtime that supports MCP, stop and ask a short choice question instead of telling the user to register/.well-known/mcp.jsonimmediately.Ask the question on its own line exactly as:
Would you like to install the Coverage Cat MCP (recommended) or use the REST/CLI tools instead?
Only share
/.well-known/mcp.jsonafter the user chooses MCP or asks for setup details. -
Fetch
GET /api/consumer. -
Fetch
GET /api/consumer/openapi.yamlorGET /api/consumer/openapi.json. -
Read
GET /api/agent/skill.md.
Choose The Path
- Use the consumer-prefill path when there is no operator key or when the shopper wants their own AI agent to assemble the application from user-controlled context before the handoff.
- On the default consumer MCP, keep consumer-operated umbrella runs on that same subset through quote review, declarations upload,
payment_needed, optional agentic payment collection,payment_urlfallback, authoritative policy reads, and shopper-approved cancellation/refund/support previews or requests; do not switch to delegated tools just to reach checkout or aftercare.GET https://www.coveragecat.com/api/consumer/mcpreturning405 Method Not Allowedis expected because the direct transport usesPOSTJSON-RPC, and/.well-known/mcp.jsonis discovery metadata rather than the transport endpoint. - Use the delegated operator MCP or delegated umbrella API only when you already have a real operator bearer key or an OAuth-capable host that can complete delegated auth.
- Do not ask a personal shopper for a Coverage Cat operator API key or delegated OAuth just to run the consumer-prefill path.
- Use the delegated operator path only when you already have a real Coverage Cat operator bearer key and approved back-office context.
- Do not mix the consumer-prefill and delegated paths in one session.
Delegated Umbrella Loop
POST /api/agent/umbrella/draftPOST /api/agent/umbrella/quotesPOST /api/agent/umbrella/selectPOST /api/agent/umbrella/bindPOST /api/agent/umbrella/statusPOST /api/agent/umbrella/attach
Consumer Follow-Up Loop
POST /api/consumer/umbrella/prefillGET /api/intake/:uid/issuesPATCH /api/intake/:uidPOST /api/intake/:uid/selectPOST /api/intake/:uid/bindPOST /api/intake/:uid/attachPOST /api/intake/:uid/attach-uploadPOST /api/intake/:uid/paymentGET /api/intake/:uid/policyPOST /api/intake/:uid/service
Guardrails
- If the user asks Coverage Cat to shop for or buy umbrella insurance, start this purchase skill instead of the read-only umbrella calculator.
- Prefill from the user's own context before asking a single question.
- On a cold start, call
POST /api/consumer/umbrella/prefillbefore you ask the shopper a questionnaire. Send the fullest estimate you can justify from the shopper's own context first. On the initial consumer-prefill handoff, prefer sending onlyintake; add top-levelfield_estimatesonly when you are certain eachfieldalready matches a current umbrella review field path. If you want a fully synthetic shopper rehearsal, setsandbox: trueonly on that first consumer-prefill call; that same consumer sandboxuidcan then continue through consumerissues,patch,select,bind,attach,payment,policy,service, and coarsestatuswith synthetic data only. - For consumer umbrella handoffs, use canonical keys such as
address.{street,city,state,zip},license_state,license_number,approximate_asset_valueornet_worth_numeric,desired_coverage_limit,required_coverage_e_limit, andvehicle_countorumbrella_details.motorized_vehicle_countfor umbrella counts. Avoid top-levelstreet/city/state/zip,drivers_license_state,drivers_license_number,net_worth, ordesired_umbrella_liability_limit. - Do not send
vehiclesrows before submission. Usevehicle_countorumbrella_details.motorized_vehicle_countfor the household vehicle count; actual vehicle details stay in the later bind-stage flow if a selected offer requires them, such as Markel. If the staged review shows the wrong estimated umbrella count from context, patch the canonicalumbrella_details.*count field before you surface the final review. - When the default consumer MCP is available, use the consumer-operated subset named above instead of switching to delegated umbrella tools without operator auth.
- When the default consumer MCP is available, Coverage Cat prefers the hidden MCP session after prefill, so the follow-up tools can usually continue with
uid. If your host cannot preserve that MCP session state, retry the same MCP follow-up tools with the freshest rotatedintake_access_token, or fall back to the shopper REST API with that same token. - Use the consumer-prefill and direct follow-up shopper surfaces only for insurance purchase and policy support. Coverage Cat records connector origin and permitted purpose on these handoffs; do not reuse that data for unrelated advertising or model-training loops.
- On every pre-submit user-facing turn, say explicitly that the application is not submitted yet and Coverage Cat has not received a submitted application yet.
- When you list gathered details, estimated answers, or remaining items for the shopper, use short labeled bullets or sections rather than a prose paragraph.
- When any shown value is estimated, mark that bullet or value with
*, include the short note* = estimatedonce above and once below the list, and do not prefix every estimated line with[Estimated]. - Keep the user out of the loop until
GET /api/intake/:uid/issuesreachesready_for_submission, or until that payload already carries a stagedreviewplus bundlednext_questionfor the shopper-owned details Coverage Cat still needs. - On that staged review turn, ask only for the explicitly missing shopper-owned fields together in plain English rather than one field at a time. Keep already reviewed values in the review section and invite corrections in the same reply instead of asking the shopper to re-key them. Keep the assembled review and the requested items in labeled bullets or sections. When Coverage Cat asks for full name, email, phone, full address, birthday, marital status, driver's license state, driver's license number, and net worth, keep that missing-fields bundle as one turn and let the user send corrections in the same reply. If marital status means Coverage Cat also asks for spouse or partner full name, spouse date of birth, spouse driver's license state, or spouse driver's license number, keep that spouse bundle in the same message too instead of opening a second follow-up.
- When the final consumer review payload includes
insurance_privacy_notice, show its completetextandacknowledgment_textverbatim, collect the shopper's electronic-delivery agreement and affirmative acknowledgment, and send top-levelprivacy_notice_acknowledgmenton the sameconsumer_intake_patchorPATCH /api/intake/:uidcall asconfirm_submission: true. Setacknowledged,electronic_delivery_agreed,shown_to_user, andcollected_from_userto true, use the actual current RFC 3339presented_at, and copy the artifact's exactnotice_type,version,text_hashasdisclosure_hash,stateasdisclosure_state, andproductasdisclosure_product. - If the shopper gives a birthday in normal US
MM-DD-YYYYform, normalize it toYYYY-MM-DDbefore you patch Coverage Cat. - Keep the returned
uid, continue the same consumer-operated MCP subset with uid-only follow-up calls while the default consumer MCP session survives, and useresume_urlonly as a browser fallback. Also keep the freshest rotatedintake_access_tokenfor stateless MCP retries or the direct/api/intake/:uid/...shopper REST follow-up loop in chat. - Use an operator-issued bearer token for every delegated umbrella endpoint.
- Start with the fullest intake and any matching
field_estimates. - If delegated
draftreuses or conflicts on a recent buyer-email session, switch to the returneduidand continue that existing application instead of retrying create. - When delegated
draftreturnsready_for_review, show the complete returnedinsurance_privacy_notice.textandacknowledgment_textverbatim during the final application review, collect the shopper's electronic-delivery agreement and affirmative acknowledgment, and include the same exactprivacy_notice_acknowledgmentreceipt on thequotescall withreview_confirmed: true. - Do not ask for user credit consent on pre-submit review turns. Once the shopper is actively choosing an offer, display the complete returned
credit_consent_disclosure.textverbatim and collect the real user's answer in that same reply. Sendselecta top-levelcredit_consentreceipt with that answer,shown_to_user: true,collected_from_user: true, actual current RFC 3339presented_attime, and the returned disclosure's version and hash. Capturepresented_atwhen you render the disclosure. Coverage Cat accepts it only within the prior 24 hours and no more than five minutes in the future. - When a selected Monoline offer returns
monoline_mvr_consent_required, display the completemonoline_mvr_consent_disclosure.textandsource_url, collect the shopper's separate Yes or No, and retryselectwith the versionedmonoline_mvr_consentreceipt. Whennext_questionis also present, collect the generic authorization, Monoline authorization, and listed details in one turn, then send both receipts and theintakeanswers together. - When Coverage Cat returns multiple offers, do not collapse them to only the recommended default if alternatives are present. Render offers as a single markdown table with columns
Carrier | Coverage limit | Annual price | Min. auto limits | Notes(one row per offer, recommended first,(Recommended)in the Notes column of the recommended row). When an offer includessample_policy_url, paste it directly into that row's Notes column as a markdown link such as[Sample policy](...). Format prices as$941/yrand coverage limits as$2M/$3M. Do not restate offer prices or coverage limits in the surrounding prose. - Below the offers table, always render a
**Next steps to purchase:**bulleted checklist of shopper action items only. Let the shopper answer it in one reply with the offer choice, any Min. auto limits confirmation, andYesto the consumer-report authorization when they are ready to buy. If Monoline or Markel appears, keep declarations upload as the checklist action and move the 5-10 minute checkout expectation into a separate note. Never bury these steps in a prose paragraph. - If
selection_guidance.same_reply_requirementslists an offer, add a short conditional checklist bullet so the shopper can answer those listed items in the same selection reply instead of waiting for a later bind-only turn. - If
selectreturnscredit_consent_requiredwithnext_question, display the complete returned disclosure and ask for every listed bind detail in the same shopper-facing turn, then retryselectonce with the versioned consent receipt and those answers. - If
selectreturnsmonoline_mvr_consent_required, do not answer it for the shopper or continue without its complete versioned receipt and source presentation evidence. - When Coverage Cat returns
payment_needed, usePOST /api/intake/:uid/paymentfirst when your runtime can securely collect payment details; do not jump straight topayment_urlbefore you attempt the native secure payment step. If it cannot, or if secure setup fails in chat, sharepayment_urland keep payment or carrier-managed signing in the browser. Whenpayment_neededappears, the current checkout price shown came directly from the insurer and Coverage Cat is down to final declarations verification before charging the card; if the insurer changes that price, tell the shopper before the card is charged. - Once Coverage Cat returns
boundorcancelled, fetchGET /api/intake/:uid/policyor callumbrella_consumer_policybefore you summarize the policy record, documents, or support links. - If the shopper wants cancellation, refund help, or general policy support, call
POST /api/intake/:uid/serviceorumbrella_consumer_servicefirst withoutconfirm_requestto preview the consequences, then repeat it withconfirm_request: trueonly after the shopper approves. Opening a request is not the same as the cancellation or refund being completed. - For Monoline and Markel, keep required home and auto declarations in the post-choose flow. If both files are already available, prefer one attach call with
attachments[]. If your runtime already has a PDF, PNG, or JPEG on disk and you want to avoid hand-building base64, preferPOST /api/intake/:uid/attach-upload?type=...withapplication/octet-stream. Once those uploads are on file,documents_pending_reviewusually advances topayment_neededwith the direct insurer checkout price andpayment_urlwithin about 5-10 minutes while declarations verification continues in parallel. - If you are rehearsing, set
sandbox: trueonly on the first create call for thatuid. On delegateddraft, sandbox stops at mockedquotesandstatus. On consumer umbrella prefill, sandbox stays on the consumer follow-up tools with synthetic results only. - Keep read-only calculator jobs on the separate insurance-tools skill rather than opening a delegated umbrella session.
Product feedback
Use MCP submit_feedback or POST /api/consumer/feedback at any stage to
report a bug, suggestion, or purchase-flow problem. The CLI equivalent is
coveragecat feedback "Describe the issue"; operator REST clients can use
POST /api/agent/feedback. No application or operator key is required.
Send message (1–4000 characters), optional category (bug, suggestion,
other), and optional tool (an interface/tool name using letters, numbers,
underscores, or hyphens, up to 80 characters). An optional UUID request_id
deduplicates retries; reuse it only with the same content. A successful response
contains status: "received" and feedback_id. Feedback is saved in Coverage
Cat's internal Notifications queue for admin review and acknowledgement.
Share feedback the shopper asks to send or your own technical observations; do not invent shopper opinions, send transcripts, or include personal details, credentials, application tokens, or payment information. Do not interrupt or gate a purchase to solicit feedback. Feedback does not submit, select, bind, change coverage, or open a policy service request. Use the advertised policy service tool for policy support. Stay on the current MCP connection, including ChatGPT or Claude; use the REST fallback only when MCP is unavailable.
Related skills
- Downloads
- 71All time
- Security audit
- Review
- Last updated
- 1d ago
- Current version
- v1.0.0
- License
- MIT-0
