Effective September 6, 2026
These terms apply to the hosted MCP Selection Lab service at https://mcp-selection-lab-production.up.railway.app.
The service provides bounded, deterministic analysis of MCP tool-selection readiness from public discovery metadata. It is not a security audit, legal or business certification, model-provider ranking, or proof of how ChatGPT, Claude, or any other model will select tools globally.
You may submit public MCP endpoints and non-sensitive test prompts for legitimate evaluation, development, interoperability, or quality-assurance purposes. Do not submit credentials, secrets, personal data, private-network endpoints, or targets you are not authorized to have contacted through ordinary public MCP discovery.
The free generated-readiness scan uses cases generated from the same tool descriptions and must not be treated as generalized Agent-routing accuracy. Caller-supplied holdout receipts are stronger only within their stated evidence boundary. If a receipt says the cases were frozen before a candidate, that status is based on the caller's attestation; Selection Lab does not independently prove when the caller created the cases.
Successful standard scans create public-by-link reports. By submitting a public MCP endpoint to the standard scan, you acknowledge that discovered public tool metadata and derived routing-readiness results may appear in that report. The caller-supplied /api/holdout endpoint does not insert its prompts or receipt into the standard public-report database.
The hosted evaluation service is designed to use MCP discovery only and not to execute target business tools, make purchases, send messages, mutate target data, or spend funds.
The public metadata scan is currently free and rate/capacity limited. Independent MCP Selection Evaluations are a separate paid, manually qualified service; payment is requested only after acceptance. We may change limits, features, availability, or pricing in the future. We may reject or stop requests that threaten service reliability or violate these terms.
Submitting an evaluation request does not create a contract for measurement, guarantee an outcome, or guarantee publication. Any public Agent Trust Lab profile requires separate review and approval. Payment does not affect scoring or evidence standards.
The service is provided on an as-available basis. Results are diagnostic evidence for the bounded test surface, not guarantees of production model behavior. You are responsible for reviewing any suggested metadata change before applying it.
We may update these terms as the product changes. Continued use after an update is subject to the then-current terms shown here.