Describe your MCP tools honestly

How to do it

Serve the card at your MCP address plus /server-card, as application/mcp-server-card+json, and list it in your AI catalog. Give it $schema, a namespaced name, version, description and remotes. Give every tool readOnlyHint, destructiveHint, idempotentHint and openWorldHint. Add outputSchema to tools that return structuredContent. Clients treat these hints as untrusted, so enforce every limit on the server too. The other side, connecting your own agent to tools, is in allow only reviewed tools.

Example

{
  "$schema": "https://static.modelcontextprotocol.io/schemas/v1/server-card.schema.json",
  "name": "com.example/bookings",
  "version": "1.2.0",
  "description": "Check free slots and book a call with Example Co.",
  "websiteUrl": "https://example.com",
  "remotes": [
    { "type": "streamable-http", "url": "https://example.com/mcp", "supportedProtocolVersions": ["2026-07-28", "2025-11-25"] }
  ]
}

{
  "name": "list_slots",
  "annotations": { "readOnlyHint": true, "destructiveHint": false, "idempotentHint": true, "openWorldHint": false }
}

How to check

The card validates against its JSON schema. An MCP inspector connects, and lists every tool with its hints.