MCP Connector

The Hedge MCP connector is a Model Context Protocol server for your brokerage’s book at Hedge. Connect it to an AI tool and the assistant can read your book: submissions, the three market categories on each one, policies, payments, appetite, programs. If you also approve write access at connect time, it can send Hedge a risk and request binds on your behalf. Everything is scoped to your own brokerage. A connected agent should show you the exact action before a write and wait for your explicit approval.

What it can read

Read access comes with every connection (the broker_mcp scope).

Submissions and markets

Status and detail of the risks in market, and get_markets: every lane in Hedge Instant Quote, Hedge Binding and Hedge Specialty with what it needs from you, the assumptions Hedge made, and released quotes.

Programs

The carrier programs your brokerage can quote, by category, and the exact application question schema for instant-quote programs.

Policies and payments

Your bound policies, their key terms, commission payments and payout status.

Appetite and market access

Which markets want a class of business, what they need to quote, and your brokerage’s own on/off preferences.

Read tools: list_my_submissions, get_submission, get_markets, get_submission_requirements, search_insureds, list_my_policies, get_policy, list_payments, get_my_book_report, check_appetite, get_market_requirements, get_market_turnaround_stats, get_market_access, list_programs, get_program_application_schema, get_carrier_quote_request, get_bind_request.

What it can do with write access

Approving the broker_submit scope on the consent screen adds the write tools. The connection can then:

• Send Hedge a risk: create_submission creates the submission and starts the run, exactly like emailing Hedge: appetite match, the clearance email to the producer, lanes opened, instant-quote markets quoted. Pass hold: true to keep the draft; finalize_submission releases a held submission. upload_document adds ACORDs, loss runs and supplements, which flow into the open lanes.
• Work carrier applications: save_carrier_answers and request_carrier_quote for instant-quote carriers.
• Request a bind: request_bind drafts the bind request for a released quote. While assumptions stand unconfirmed it returns code: assumptions_unconfirmed with the list; the agent shows you each statement and calls again with attest_assumptions only after you agree. get_bind_request tracks it.
• Manage market preferences: set_market_access turns a market, one of its programs, or a single line of business on or off for new submissions (brokerage admins).

Replying on the thread and answering the batch of outstanding asks are REST (and CLI) surfaces today: POST /broker/submissions/{id}/thread and POST .../outstanding-requirements/answers, or hedge reply and hedge answer-asks.

Note

Two switches on your side control writes: the broker_submit scope you approve at connect time, and the Agent write access setting under Settings → Connected apps in the broker portal. Turn the setting off and every connected tool drops back to read-only immediately. Toggling market access additionally requires an admin account, matching the portal.

Recommended action sequence

1. check_appetite if you want a read on the class before sending anything.
2. create_submission with the insured, narrative, lines and the producer’s contact_email. This starts the run. Use hold: true when you want to attach documents and review requirements before Hedge starts; the agent should say plainly which it is doing.
3. get_markets once matching lands (usually within about five minutes). Show the three categories, each lane’s needs_from_you, and the assumptions with confirmed: false.
4. Resolve the asks: upload_document for files, save_carrier_answers for carrier questions, and a reply on the thread (hedge reply or the REST endpoint) for anything Hedge asked by email.
5. request_bind for the quote you chose. When it returns assumptions_unconfirmed, read every statement to the broker; only after they agree, call it again with attest_assumptions: true (or the confirmed keys). The confirmation is recorded under the broker’s name.
6. Contingency documents and the final submit are REST/CLI (hedge bind-upload, hedge bind-submit); get_bind_request reports status through bound.

Never put insured, producer, agency, credential, file, health, or other private submission data in a public URL or search query.

Connect it

The connector works with any MCP-capable client, including Claude, ChatGPT, Cursor, Codex and Claude Code.

1. Copy the connector URL. Find the Hedge connector URL in the broker portal under Settings → Integrations.

2. Add it to your AI tool. Paste the URL into your client’s connector or MCP server settings:

   • Claude and Claude Code: add it as a custom connector.
   • ChatGPT: add it as a connector.
   • Cursor: add it as an MCP server.
   • Codex: the Integrations pane shows the config.toml entry (via the mcp-remote bridge).

3. Sign in and approve. The tool opens the broker portal. Sign in and approve the connection. The consent screen shows exactly what the tool will be able to do; write access is a separate line you can decline for a read-only connection.

4. Ask away. “Which of my submissions are waiting on me”, “what did Hedge assume on the Coterie quote for Acme”, “send Hedge this roofing risk and show me the markets”, “request a bind on the Kinsale quote once I’ve confirmed the assumptions”.

Manage and revoke access

Every connected tool shows up as a connected app. Manage or revoke them any time under Settings → Connected apps in the broker portal. You can revoke a single connection or disconnect everything at once. A revoked tool has to go back through the approval screen if it reconnects.

The connector authenticates with OAuth 2.1, the same model the CLI and API use; machine credentials (API keys) are not accepted by the MCP write tools. See Authentication for details.