Skip to main content

Integrations

Everdict is driven by people through the web app, and by machines through MCP. These pages are the machine half.

  • MCP — the agent-facing surface, at parity with the HTTP API by construction
  • Claude Code plugin — two commands: the tools and the domain context
  • Running Codex — Codex as the agent under test, via a declarative harness
  • Desktop app — the shell that carries a runner on your machine

One distinction worth getting straight before you read further. An agent can be the driver (it operates Everdict) or the subject (Everdict evaluates it). Claude Code can be either, and the two are independent — you can drive Everdict from a Claude Code session while evaluating something else entirely.

Driving is MCP. Being evaluated is a harness.