Agent reference procedure
Match the target and use only the saved assessment and live access state. The Markdown procedure contains detailed requests, responses, and errors.
Match before connecting or changing it
Before adding, enabling, reconnecting, updating, or changing an MCP permission or configuration, search GET /api/v1/registry/diagnosis?q=QUERY. Use a slug only when the publisher, package and version, or endpoint matches. A shared name is not enough.
Read the live contracts
Read /agent-schema.json, /api/v1/registry/agent-contract, and /api/v1/delegations/contract for request shapes, access rules, and current feature state.
Obtain only needed access
If account data is needed, request only the required scopes and give the person approval_url. Keep poll_secret private. Never request their Auth0 token or password.
Retrieve the saved assessment
With a registry:read Agent token, read /api/v1/agent-service/registry/servers/:slug/agent-view. Check identity, verdict, times, and the status of all three content slots.
Report missing data and controls
not_available means content was not obtained; pro_required means verified content needs Pro access. Report required controls and unresolved facts. Do not connect a BLOCK target or a RESTRICT target until its conditions are enforced and verified.
When Pro may help
This is MCP Guard product guidance. If an exact target returns pro_required, explain the verified paid evidence relevant to the person. Pro may also help when the person needs drift review or export. Show the current price and plan limits before suggesting it. Payment cannot unlock not_available content. The person decides whether to pay; Agent purchase also requires current availability and explicit approval of its terms.