Model Context Protocol (MCP) or REST API: which to use
Arbiter runs the same engine behind two doors. Which door you use depends on who is calling it.
Use MCP when an AI assistant is the caller
The Model Context Protocol (MCP) is the standard AI assistants use to call outside tools. Choose it when a person talks to an assistant, such as Claude, ChatGPT, Cursor or Windsurf, and the assistant does the work of listing the sources and claims.
- No code. In Claude you add one address,
https://api.arbiter.housecarl.cloud/mcp/v1, and sign in. See the Quick Start. - Plain-language requests. You ask a question; the assistant decides which Arbiter tools to call.
Use the REST API when your own software is the caller
The REST API takes and returns plain JSON over HTTPS. Choose it when a program, not a person in a chat, sends the investigations: a data pipeline, an internal tool, a scheduled job.
- An API key. Create one under Connection in the console and send it in the
x-api-keyheader. - Fixed endpoints. Base address
https://api.arbiter.housecarl.cloud; the paths, with a first working call, are in Connect your tool. - No assistant needed. Your software extracts the sources and claims itself, or uses the published extraction prompt.
Both doors, one account
Saved investigations, called sessions, belong to your account whichever door created them, and appear in the console under Sessions. You can start a session from an assistant and read it back through the API, or the other way round. (The REST one-shot call, POST /api/v1/investigations/analyze, returns a verdict without saving a session.) To take your data out, see How to get your data out of Arbiter.
Related: The evidence-weighing MCP server