Common questions about the connector.

Questions and answers.

Does it require my customers to install something?

No. The connector is per-user and per-firm: the public adjuster signs into Claude Desktop or ChatGPT and authorizes claimOS once. The insured client never installs anything; they continue to interact with the firm through the existing client portal, email, and phone.

Can my paralegal, intake coordinator, or junior adjuster use it?

Yes. Each user authorizes the connector with their own claimOS credentials, and the scopes inside the connector match the permissions they already have inside claimOS. A paralegal who can only see assigned claims in the web app sees the same subset through the connector. The connector does not widen access.

Is data sent to Anthropic, OpenAI, or other AI providers?

Only what the user explicitly asks the assistant to fetch. The connector returns claim data to the assistant on the user's behalf, the assistant uses that data to compose an answer, and the answer is shown to the user. We do not pre-load data, do not send background syncs, and do not maintain a copy of your data outside claimOS. Anthropic's and OpenAI's data-handling terms govern what they do with the assistant traffic; check the privacy and zero-retention options on those providers' admin consoles.

What happens if my claimOS subscription lapses?

Tokens stop working immediately. The next request from the assistant returns 401, and the user sees an authorization error in the chat. After re-subscribing, re-add the connector and re-authorize; nothing about your data needs to be re-imported.

How do I revoke access?

Open Admin → Settings → Connected Apps inside claimOS. Each authorized assistant is listed with the scopes granted, when access was granted, and when it was last used. Revoke is a single click; the next request returns 401 within seconds. Revocation preserves the audit trail.

What if Claude or ChatGPT goes down?

claimOS keeps working. The connector is additive, not required: the web app, the iOS app, and every internal workflow function whether or not an external assistant is reachable. Outages on the AI provider side affect what the assistant can answer, not what claimOS can do.

Can I gate write access?

Yes. Issue a connection without claim:write and the assistant gets a strict read-only credential. Calls to log_task or log_note return an insufficient_scope error. Many firms start read-only and add write later once they trust the assistant's behavior.

How is the connector priced?

Included with any active claimOS subscription. No additional cost, no per-token fee, no extra seat charge. The assistant's own subscription (Claude Pro, Claude Team, ChatGPT Team, etc.) is billed separately by the AI provider.

Which AI assistants does it work with?

Anything that speaks the Model Context Protocol (MCP) over HTTP and supports OAuth 2.0 Bearer tokens. Today that includes Claude Desktop, ChatGPT (Team, Business, Enterprise, and Education plans for custom connectors), and Cursor. New MCP-compatible clients work without claimOS doing anything; the protocol is open.

Can I write my own MCP client against claimOS?

Yes. The OAuth metadata document at /oauth/.well-known/oauth-authorization-server has the authorize and token URLs; the MCP endpoint is /api/mcp. The wire format is JSON-RPC 2.0; see the tool reference for the request shape. Standard MCP libraries like @modelcontextprotocol/sdk work without modification.

What's the rate limit?

Per-user limits match the rest of the claimOS API. Most assistant workflows are bursty (several tool calls per question, then idle); the limits are tuned for that. If you're building a custom client that issues sustained traffic, email hello@claimos.net before you go to production.

Does the connector see my password?

No. OAuth specifically exists to keep credentials out of third-party apps. The assistant never sees your claimOS password. The consent screen is served by app.claimos.net; you sign in there in your browser and approve the scopes; the assistant only ever sees the resulting token.

Can I see what the connector has been doing on my account?

Yes. Connected Apps shows last-used time per assistant, and the audit trail records each authorization, refresh, and revocation. Within claims, write actions (log_task, log_note) are attributed to the connector with the client id of the assistant that made the call.

Still have a question?

Email hello@claimos.net. Tell us which assistant you're using, what scopes you've granted, and what you're trying to do. For account-level revocation, you can act immediately on the Connected Apps page.