# Install CarsXE — agent prompt

Paste the prompt below into your agent (Claude, ChatGPT, Cursor, OpenClaw…) and it will connect itself to the remote MCP server.

## Install prompt

```text
Use CarsXE as a remote MCP server over HTTP/streamable-http at https://api.mcp.ai/p_carsxe?ms=1787293620000. Don't treat it as local/stdio, don't run npx/node to "spin up" a server, and don't assume that just registering the URL in a config is enough — the goal is to actually connect and call the tools.

Authentication (client without browser OAuth): (1) list the tools — you should see show_version, report_bug and authenticate; (2) call authenticate with no arguments → it returns a login link (/agent-auth); (3) show that link to the user and ask them to open it, log in at mcp.ai, and copy the access token (Bearer JWT) shown on the page; (4) when they paste the token back, call authenticate again with { "token": "<jwt_pasted_by_user>" }.

Once identity is confirmed, call connect → if it returns an authorization link, show it to the user so they can open it and authorize access. Once connected, use the server's read-only tools (e.g. carsxe_check_lien_and_theft, carsxe_decode_license_plate, carsxe_decode_obd_code, carsxe_get_market_value, carsxe_get_recall_batch, carsxe_get_vehicle_history).

Be operational: actually try the connection and run the whole flow; don't stop at "I added the URL" and don't say "it's not possible" without first trying authenticate and connect.
```

## MCP server URL
Remote endpoint (HTTP, streamable): `https://api.mcp.ai/p_carsxe?ms=1787293620000`

## More
- Page: https://mcp.ai/carsxe
- How to use (skill.md): https://mcp.ai/carsxe/skill.md
- Agent spec (llms.txt): https://mcp.ai/carsxe/llms.txt