AgentClash for ChatGPT

Turn refund failures into meaningful regression tests, run your Python agent locally, and compare baseline and candidate reports. The four public tools require no AgentClash account.

The first supported template is a Python callable with injected refund fixtures. Compilation validates test files; it does not execute or improve an agent. Local evaluation measures only the selected target under the recorded fixtures.

MCP endpoint: https://mcp.agentclash.dev/mcp

Download Python SDK 0.2.0a1 (verified wheel; Linux execution tested). SHA256: 1e88b0937555e9233ed9c1562fb4f6095d7a2d0b371b5bbd1d9f5f1cc3a46ca7.

Source and installation guidance: AgentClash repository. The hosted integration is an engineering candidate pending OpenAI review and publication.