---
title: "Coverage Cat Connector Data Handling"
description: "Coverage Cat connector privacy notice: collected and returned data, authenticated website details, recipients, permitted uses, retention, revocation and deletion requests."
canonical: "https://www.coveragecat.com/developers/data-handling"
last-updated: "2026-11-04"
---

# Coverage Cat Connector Data Handling

## Scope

This notice applies to data submitted through Coverage Cat's agent and connector purchase surfaces:

- the hosted ChatGPT connector at `/api/chatgpt/mcp`
- the hosted Claude connector at `/api/claude/mcp`
- `/api/consumer`, `/api/consumer/openapi.yaml`, and the default consumer MCP at `/.well-known/mcp.json`
- direct follow-up requests under `/api/intake/{uid}`
- delegated operator APIs under `/api/agent/*`
- the delegated operator MCP at `/api/agent/mcp`
- related browser continuations, secure payment UI handoffs, and policy-support follow-up opened from those flows

## Precedence for connector data

For data submitted through those connector surfaces, this notice controls if it conflicts with Coverage Cat's broader website privacy policy, marketing-site terms, or similar general-purpose documents.

## Data collected and returned by ChatGPT and Claude tools

The hosted connectors return only the information needed for the requested calculation, application step or policy support. The same limits apply to nested tool fields, text copies of results and application-card data.

| Category | What we receive and why | What the connected assistant receives |
| --- | --- | --- |
| General research and optional calculations | A state for general price ranges. Optional planning tools use supplied net worth, protected assets, annual income, a planning horizon, underlying liability limits, premiums, switching costs and whether coverage is comparable. | State-level price ranges, or the supplied numeric calculation inputs, arithmetic results, assumptions and limitations. These tools do not require an account or create an application. Financial planning inputs are separate from the coarse estimate application. |
| Property insurance advice | Your explicitly confirmed eligible state, property type, question and optional non-identifying context. Account authentication and usage limits apply. Do not send identifiers, credentials or policy documents. | An educational answer, coverage checks, next steps, follow-up questions, cited knowledge-base sources when available, and the limits of the answer. This tool does not access policy documents or change coverage. |
| Account connection | Your email address on the Coverage Cat sign-in page; verification and OAuth records; account and client identifiers, granted permissions, access and refresh tokens, and browser session cookies. These establish and protect account access. | An authentication challenge when connection is needed. Email addresses, internal account identifiers, credentials, access/refresh tokens and session cookies are not returned in tool results. Credentials travel through the authentication channels. |
| Application references and choices | An application reference (`uid`), an offer selection reference (`selection_token`), approvals, a separately answered credit-consent question, and optional request-deduplication keys. These associate the requested action with the right application or offer and avoid duplicate writes. | Application references when resuming or listing accessible applications, and selection references when choosing an offer. These references do not replace account authorization. Request-deduplication keys and internal audit identifiers are not returned. |
| Coarse estimate profile | State, requested coverage and underlying auto/home liability limits, insurance/limit confirmations, vehicle/property/watercraft counts, acreage and other aggregate exposure counts, household size, driver age-band counts and an optional new-driver count. Inferred answers may include an allowed field name, source, confidence and confirmation flag. | The saved coarse review, missing coarse questions, inferred-answer markers, preliminary estimates and offer terms needed to compare or correct that application. No names, contact details, exact ages, birth dates, individual histories or declarations are returned. |
| Application, payment and policy status | The selected application's saved workflow, insurer and policy records. | Carrier, premium, coverage limits, effective dates, requirements and coarse completion flags for private details, declarations, payment setup and policy issuance. Policy numbers are omitted unless you explicitly ask for your policy number and the policy lookup requests it. Website links for details, checkout or policy downloads require website authentication and contain no sign-in or document-download bearer token. |
| Policy support requests | Request type, reason or details you choose to supply, requested effective date and explicit approval to submit. We use these to handle that support, cancellation or refund request. Do not include credentials, card information or unnecessary personal details. | The current request's type, status, effective date, approval requirement and coverage/billing consequences. We do not echo the reason or details, prior request history, internal request identifiers or event timestamps. |

## Private information collected on Coverage Cat

After offer selection, the authenticated website collects the information needed for underwriting and purchase: names and contact details; street addresses; exact ages or birth dates; marital and household information; occupation and financial information; driver licence details, driving and claims history; vehicle details including VINs when required; and insurer-specific eligibility answers. Household members' information is collected only as required for the application.

Declarations and policy documents can contain these personal details, policy numbers and original filenames. Uploads and document downloads take place on Coverage Cat. Hosted tool results do not contain document contents, original filenames or signed document-download credentials. They report required document types and coarse upload completion instead. Payment credentials are entered in the hosted payment interface and processed by the payment provider; the assistant receives only the relevant payment status.

Public consumer and delegated operator APIs have their own published input/output schemas and may receive or return the application, document and policy information required by those workflows. The narrower ChatGPT/Claude tool contract above applies to the hosted endpoints, not to every public API client. All these records remain subject to this notice's purpose and sharing restrictions.

## Operational and diagnostic data

We process account/application associations, client and connector origin, permission and consent records, operation/request identifiers, timestamps, security events, and technical request data such as IP address and user agent to authenticate requests, prevent duplicate actions, diagnose failures, attribute changes and meet insurance recordkeeping obligations. These server records are not added to hosted tool results as debug objects or hidden user metadata.

Protocol request IDs, authentication challenges and static UI/resource metadata support transport, sign-in and application-card rendering. They do not expose the underlying account, session, trace or audit records. Cached results pass the current response restrictions before being returned.

## Recipients

Your connected platform (OpenAI for ChatGPT or Anthropic for Claude) receives the tool inputs sent from your conversation and the limited results described above so it can answer your request. That platform handles your conversation under its own privacy terms. Entering information directly on Coverage Cat does not by itself return that information to the assistant.

Property advice questions, optional context and retrieved knowledge-base excerpts are sent to OpenAI under Coverage Cat's Zero Data Retention agreement. We redact email addresses and phone numbers before retrieval and generation, filter question/context from request logs, and do not create stored assistant conversations. Account usage metadata enforces limits across connectors; REST idempotency responses may be cached for up to 24 hours. The connected platform's conversation retention is governed by that platform.

Coverage Cat shares information needed to quote, underwrite, issue or service the requested coverage with the relevant insurers and insurance service providers. Providers supporting hosting, authentication, document processing, payments, communications and support process information needed for those functions. Disclosure may also be required to meet legal, regulatory or fraud-prevention obligations. These uses remain limited to the purposes below.

## Data minimization and vendor retention

To protect our customers, Coverage Cat strongly prefers business tools, services and configurations that enforce zero data retention (ZDR) within their agreed scope and otherwise minimize data retention. We seek to limit the information shared, access granted and time records are kept to what the permitted workflow needs.

ZDR applies to submitted content and generated outputs within the provider's covered service and features. We assess the contractual commitments, technical settings, retained metadata and legal or security exceptions before relying on it. A promise not to train models on data does not by itself establish ZDR. A connected assistant platform or independent operator handles its own copies under its own terms; Coverage Cat's OpenAI agreement does not extend to those copies.

We may use providers without ZDR when needed for a permitted workflow, subject to appropriate contractual and technical safeguards and the purpose restrictions below. Coverage Cat retains account, insurance and operational records as needed to provide the requested service and meet the obligations described below.

## Permitted purpose

Coverage Cat uses connector-submitted data only for:

- insurance purchase workflows
- quote, bind, payment-method, document, and policy-support steps tied to that workflow
- support, fraud-prevention, audit, security, compliance, and legally required insurance recordkeeping that directly relates to servicing that workflow

## Prohibited uses

Coverage Cat does not use connector-submitted data, or datasets derived from it, for:

- unrelated advertising
- remarketing unrelated to servicing the requested insurance workflow
- model training or evaluation unrelated to serving this workflow

## Purpose labeling and origin tracking

Coverage Cat records connector origin and permitted purpose on agent-driven intake touches. Current connector-origin labels include:

- `consumer_mcp`
- `chatgpt_mcp`
- `claude_mcp`
- `consumer_rest`
- `operator_mcp`
- `operator_rest`

Agents and internal services that store or replay connector data should preserve that same scope.

## Revocation, deletion, and retention

- You can disconnect Coverage Cat in your chat platform. Hosted consumer OAuth tokens can be revoked at `/consumer/oauth/revoke`; contact us if you need help revoking access. Revocation stops that authorization but does not erase a conversation already held by the platform or cancel an insurance policy.
- Operator OAuth access can be revoked at `/oauth2/revoke`.
- Consumer direct-follow-up credentials are short-lived and rotate as Coverage Cat returns fresh handoff state.
- Contact [Coverage Cat support](/contact) or [support@coveragecat.com](mailto:support@coveragecat.com) to request access, correction or deletion of your connector data, or help with revocation. We verify account ownership before acting on these requests.
- We retain account, application, document, policy, support and operational records for as long as needed to provide and support the requested service and meet insurance, billing, fraud-prevention, tax or legal obligations. Disconnecting does not automatically delete these records. Required retained records stay limited to the purchase-and-policy-support scope above.

## Handling expectations

- Do not send connector data into unrelated advertising systems or unrelated model-training loops.
- Do not treat de-identification or aggregation as an automatic exception to the purpose restriction.
- Keep secure payment or document-upload secrets in the secure UI or authenticated connector channel, not in model-visible transcripts or prompts.

## Public machine-readable references

- Discovery: `/api/agent` and `/api/consumer`
- OpenAPI: `/api/agent/openapi.yaml` and `/api/consumer/openapi.yaml`
- Contact: `/contact`
- Hosted consumer auth revocation: `/consumer/oauth/revoke`
- Operator auth revocation: `/oauth2/revoke`

## Multi-language starter samples

Coverage Cat's shopper-safe consumer discovery document is the fastest starting point for a personal shopping agent. From there, call the consumer-prefill flow instead of sending the shopper straight to `/umbrella`.

```sh
curl https://www.coveragecat.com/api/consumer
curl -X POST https://www.coveragecat.com/api/consumer/umbrella/prefill \
  -H "Content-Type: application/json" \
  -H "Coverage-Cat-API-Version: v1" \
  -H "Idempotency-Key: demo-consumer-prefill-1" \
  -d '{"credit_consent_pending":true,"intake":{"full_name":"Taylor Example","email":"taylor@example.test","address":{"street":"123 Main St","city":"San Francisco","state":"CA","zip":"94107"}}}'
```

```javascript
const discovery = await fetch("https://www.coveragecat.com/api/consumer").then((response) => response.json());
console.log(discovery.preferred_openapi_url);
```

```python
import requests

discovery = requests.get("https://www.coveragecat.com/api/consumer", timeout=30).json()
print(discovery["preferred_openapi_url"])
```