Privacy — AgentClash anonymous evaluation tools

Effective October 28, 2026. This notice covers the anonymous MCP service on mcp.agentclash.dev, operated under the AgentClash project.

What you deliberately send

Tools process the suite specifications, test prompts, fixtures, exact bundle files, and baseline/candidate evaluation reports that you choose to submit. These may contain personal information if you include it. Sanitize customer text, identifiers, traces, credentials, and other sensitive information before submission.

Use and storage

We use supplied artifacts to validate or compile regression files and compare reports, then return the result to the requesting client. The application processes these inputs in memory for the request; it does not create a persistent database record of suites or reports and does not log their bodies. It does not use these submissions to train models, call an AI provider, fetch user URLs, or run uploaded code.

The service is hosted on Vercel. Vercel processes network and request metadata, which can include IP address, request path, timestamps, response status, and runtime diagnostics, to deliver and secure the service. Infrastructure metadata is subject to Vercel’s operational retention controls and privacy policy; it is not the same as application storage. We do not promise that hosting metadata is immediately deleted. No analytics or advertising scripts are included in these pages.

Local execution and other services

Your Python target and any provider calls run in your own environment. Provider credentials stay there and are not required by this MCP service. ChatGPT and any provider you independently use process data under their own terms.

Questions and requests

Use the support channels for privacy questions or access/deletion requests. Do not post private reports or secrets in public issues. Since this anonymous application stores no suite/report records or accounts, it cannot look up a private report history or another user’s workspace.