home-auto-insurance-quotes

Quotor — an agent skill that gets real home & auto insurance quotes, and starts a binding, through a network of licensed independent agencies.

This repository packages a portable Agent Skill that any AI agent or assistant (Claude, ChatGPT, Hermes, OpenClaw, Codex-style agents, and other MCP-capable runtimes) can pull in to quote and help bind home, auto, and bundled insurance. The skill is the front door; a hardened MCP server is the infrastructure behind it.

Coverage: Texas today (home, auto, home+auto bundles; renters & other lines via lead capture), expanding nationally. The skill always checks eligibility per state and product before quoting.

What it does

• Real quotes, not estimates. Returns live multi-carrier rates for the customer's state and product.
• Two integration styles:
  • Conversational — an assistant does a quick intake in chat and reads back quotes.
  • Agentic — a system connects the MCP server directly and runs the full quote-to-bind flow.
• Quote → bind handoff. When the customer is ready, the skill starts a binding by routing them to a licensed agency with a secure link. A licensed human finalizes the policy — the network orchestrates the shopping; the agency owns the relationship.

How it's structured

home-auto-insurance-quotes/
├── SKILL.md                      # the skill itself (agentskills.io spec)
├── README.md                     # this file (for humans browsing the repo)
└── references/
    └── mcp-connection.md         # endpoint, auth (none required), tool schemas, raw examples

• SKILL.md is what agents load. It documents the two usage pathways, the tool set and call order, intake guidance, worked examples, and the MCP connection details.
• references/mcp-connection.md is the technical reference for the MCP endpoint: URL, authentication, context headers, per-tool schemas, and copy-paste JSON-RPC examples.

Connecting to the MCP

• Endpoint: https://mcp.quotor.ai (Streamable HTTP, JSON-RPC 2.0).
• Auth: none required. All tools work anonymously — connect and quote, no sign-up, no key (anonymous traffic is rate-limited and spend-capped server-side). End customers never need a key.
• Optional partner key (instant, self-serve): integrating at volume? POST <endpoint>/keys with { "partner_label": "...", "email": "..." } → returns an api_key shown once. Send it as Authorization: Bearer <api_key> to give your integration its own rate/spend caps and attribution instead of sharing the anonymous pool.

Add it to your MCP client (claude.ai custom connectors, Claude Desktop, Cursor, VS Code, or any Streamable-HTTP MCP client) — no headers needed:

{
  "mcpServers": {
    "quotor": {
      "type": "http",
      "url": "https://mcp.quotor.ai"
    }
  }
}

The machine-readable manifest is server.json (official MCP registry schema) — that's what client tooling and directory crawlers read.

See references/mcp-connection.md for the full handshake.

Security & privacy posture

• Quote-only scope. The MCP cannot reach carrier portal credentials, cannot service or cancel existing policies, and cannot finalize a bind without a licensed human. A bind request is a routed handoff, not an automated purchase.
• Rate- and spend-capped: anonymous traffic draws per-IP and per-quote budgets under a hard daily spend ceiling; optional partner keys carry their own caps — no single caller can run up cost.
• Every call logged; anomalous usage (auth-failure spikes, off-hours surges) is monitored.
• Customer data is protected: quotes create prospect records and never overwrite existing customers; a returning customer's current premium is only ever returned to an authenticated partner, never on an anonymous call.

Status

Live. The MCP is deployed in production — rate and spend caps and full audit logging are in force; API keys are optional (partner attribution + dedicated caps). Coverage is Texas today (home, auto, bundles; other lines via lead capture), expanding nationally; always call check_eligibility first.

License

Apache-2.0.