Carrierscore

What the carrierscore MCP server does

The carrierscore MCP server gives freight-broker and AI booking workflows access to CarrierScore’s analysis of public FMCSA and US DOT safety data. It addresses two related questions: how risky a carrier appears under the service’s methodology, and what information was available at a particular point in time.

The available tools are:

• carrier_lookup verifies a US DOT number and returns identity details such as legal name, DBA, operating status, FMCSA safety rating, fleet size, address, and registration dates. Use this before relying on a score to confirm that the DOT number matches the rate confirmation.
• carrier_score returns a 0–100 CarrierScore risk result, where higher values indicate greater risk. The response includes component-level detail, a data-sufficiency indicator, and hard flags for conditions such as an active out-of-service order, missing active insurance, or a possible reincarnated-carrier link.
• montgomery_file creates a timestamped carrier-selection evidence report for the booking decision. Free-tier responses are text; paid tiers provide JSON.
• monitor_carriers evaluates up to 100 DOT numbers in one call, returning score summaries and flags for candidate screening or roster review.

How it works

CarrierScore builds longitudinal results from daily archives covering census information, inspections, crashes, out-of-service orders, operating authority, and insurance filings. Its scores use population percentiles together with hard flags. Evidence reports preserve the carrier-selection information associated with the time of booking, including the report’s timestamp.

The service is exposed through Model Context Protocol over Streamable HTTP. The hosted endpoint is https://mcp.carrierscore.io/mcp, so an MCP client can connect without installing the repository. Keyless access uses a rate-limited free tier. Paid access adds higher volume, carrier-list monitoring, and structured JSON evidence reports.

Setup and configuration

For the hosted service, configure an MCP client with the endpoint URL. The README provides connection examples for Claude Code, Claude Desktop or claude.ai, Cursor, and generic Streamable HTTP clients.

Self-hosting requires Node.js project installation and a build before starting the process:

npm install
npm run build
CARRIERSCORE_API_URL=https://api.carrierscore.io npm start

The local MCP endpoint listens on POST /mcp and defaults to port 8322. CARRIERSCORE_API_URL selects the CarrierScore REST API base URL; CARRIERSCORE_API_KEY can authenticate paid-tier requests and is optional for the rate-limited free tier; PORT changes the MCP listener port. The repository implementation is a thin Streamable HTTP client over the REST API rather than the scoring data service itself.

Limitations and notes

The carrierscore MCP server reports statistical indicators derived from public FMCSA data. Its results are not safety-fitness determinations, are not endorsed by FMCSA, and should not be treated as a replacement for an official safety rating or a broker’s own review.

Scores and evidence reports include the service disclaimer. Retain that disclaimer when storing or quoting results. Free access has rate limits and returns text evidence reports; paid tiers are needed for structured JSON evidence reports and additional volume-related features. The batch monitoring tool accepts no more than 100 DOT numbers per call.