Skip to main content
The AI Insurance MCP connector lets you work with your AI Insurance data from Claude — in a conversation, you can ask Claude to look up a submission, summarize a policy’s endorsement history, bind a prepared quote, create a task, draft a note on a claim, or export your exposures to a table, and Claude does it against your live AI Insurance account. It’s built on the Model Context Protocol (MCP), an open standard for connecting AI assistants to external systems. You add AI Insurance as a custom connector in Claude.ai or Claude Desktop, sign in once with your normal AI Insurance account, and Claude gains a set of tools for reading and writing your data — acting as you, with exactly the permissions your user has.
The MCP connector has not been released yet. This page documents the connector ahead of its launch — the connector URL below will not accept connections until the rollout completes. Watch the changelog for the release announcement.
The MCP connector is in beta. Once released, it may need to be enabled for your company before the connector will accept connections. If you can’t connect, reach out via the support portal or your AI Insurance contact to get access.

​
Adding the connector

The connector URL is:
1

Open your Claude connector settings

In Claude.ai, go to Settings → Connectors and choose Add custom connector. In Claude Desktop, the same option lives under Settings → Connectors. (Custom connectors require a Claude plan that supports them; see Claude’s own documentation for plan availability.)
2

Enter the connector URL

Give the connector a name (for example, “AI Insurance”) and paste the URL https://mcp.go.aiinsurance.io/mcp. No API key or other credentials are entered here — authentication happens in the next step.
3

Sign in with your AI Insurance account

The first time you connect, a browser window opens asking you to sign in to AI Insurance. Use the same account you use for the web app at go.aiinsurance.io. Once you approve, Claude is connected — you won’t need to sign in again unless your session expires or you disconnect.
4

Try it out

Start a conversation and ask something like “Which AI Insurance companies can I access?” or “Find the submission for Acme Manufacturing and summarize where it stands.” A good first prompt is to ask Claude to call the connector’s get_started tool — it teaches Claude how your data model works before it starts reading and writing.

​
How access and permissions work

The connector acts on your behalf:
  • You sign in as yourself. There are no shared credentials and no API keys — each user connects with their own AI Insurance login.
  • Claude can do what you can do — no more. Every request is checked against your user’s role and permissions, per company, exactly as in the web app and the REST API. If your role can’t delete quotes, Claude can’t either.
  • Multi-company users pick a company. If your account spans several companies, Claude first lists the companies you can act on and works within the one you choose.
  • Everything is logged. Actions taken through the connector are recorded in the same audit log as other API activity.
Claude’s writes are real. When you ask Claude to create, update, or delete a record through the connector, it changes your live data — the same as making the edit in the app. Claude will treat a clear, specific request as confirmation to proceed, so be deliberate when asking for deletions or bulk changes.

​
What Claude can do

Once connected, Claude has tools covering your core records, policies, tasks, and notes.

​
Orientation

​
Records (submissions, quotes, events, exposures, people, organizations)

​
Binding quotes

​
Policies (read-only)

Policies are managed through their transaction lifecycle (bind, endorse, cancel, reinstate, renew), so the policy tools themselves are read-only — Claude can analyze your book, and the one way it changes a policy is by binding a prepared quote with bind_quote above.

​
Tasks and notes

​
The configuration connector

Alongside the data connector, there is a second, separate connector for editing your company’s configuration — the fields, option sets, custom objects, and page layouts that define how your app works (see Configuration Management for the equivalent REST surface). Its URL is:
Add it the same way as the data connector. Because both use the same sign-in, you won’t be asked to log in a second time. Its tools let Claude inspect your configuration (describe_config, describe_entity, describe_fields, describe_field_types, describe_option_set, describe_cards, describe_card, describe_placement_options), change it with precise, single-purpose operations (add_field, update_field, delete_field, create_option_set, add_option_set_option, update_option_set_option, delete_option_set, delete_option_set_option, create_custom_object, delete_custom_object, add_card, update_card, delete_card, add_field_to_card, update_field_on_card, remove_field_from_card, relayout_card, or apply_changes for a multi-part change committed atomically), and bootstrap a new company from starter modules (list_starter_modules, seed_configuration). Configuration changes are higher-stakes than data entry, so this connector is built around a validate-first workflow: every change supports a dry run that checks the full result without committing anything, and Claude is instructed to confirm destructive changes with you before applying them.
The configuration connector is an earlier-stage beta than the data connector and is enabled separately. If the URL returns “not found”, it isn’t enabled for your environment yet — contact support to request access.

​
Tips for good results

  • Start with get_started. Asking Claude to call it first gives much better results — it learns your entity types and how option values work.
  • Be specific about records. “Delete the duplicate exposure for 123 Main St” works better than “clean up the exposures” — Claude confirms with you when a request is ambiguous or implies a bulk change.
  • Let Claude read the schema before writing. For creates and updates, Claude reads the configured fields first so the data it writes matches your setup. If a value is rejected, the error names the exact field so it can fix and retry.

​
Troubleshooting