Claim verification API, MCP, and setup
Start where it suits you. Short on time? Quick Start is the two-minute connector setup. How it works explains, in plain terms, how Arbiter turns conflicting sources into a verdict you can defend — and walks a complete financial-fraud investigation end to end. Use it from Claude gets you running in five minutes with no code and no card. Use it from Google Antigravity covers the one config difference that trips people up (Antigravity uses serverUrl, not url). Connect your tool is the developer path: REST and MCP, with a quickstart call.
For agents: machine-readable context lives at /llms.txt (concise) and /llms-full.txt (full). To turn raw source text into a ready-to-submit investigation payload, hand a coding agent the extraction prompt at /extract.txt.
-
Extract claims from any text
A prompt that turns source text, such as an article, transcript, court filing, earnings call or memo, into an Arbiter investigation in JSON.
-
How it works, with a worked example
What Arbiter does: which sources to trust, which claims are out of date and what the evidence supports. Then a financial-fraud investigation, step by step.
-
Use it from Claude (no code)
The quickest way to start: add Arbiter to Claude as a custom connector, then ask questions in plain language. It works on the free plan and on your phone.
-
Use it from Google Antigravity
Connect Arbiter to Google Antigravity as a Model Context Protocol (MCP) server. Antigravity uses serverUrl, not url, and authenticates with an API key.
-
Connect your tool (API & MCP)
How to call Arbiter from your own software: authentication, a sample curl call, the core endpoints, operating limits and the full OpenAPI reference.
-
Source types & reliability rubric
All eleven source_type values with their reliability anchors, rationale, and temporal decay multipliers; what happens on an invalid value; and how to model a source that isn't in the list.