Reading a Server Manifest Like a Reviewer
A server's discover result, its tool list, and its registry entry are the only things you know about it before you call anything, so read them the way you read a contract, because an unread default becomes a promise you did not mean to keep. Read a server/discover result, a tools/list page, and a registry server.json as the three documents that describe a server before any call is made. Apply the tool annotation defaults (readOnlyHint false, destructiveHint true, idempotentHint false, openWorldHint true) to see what an omitted annotation actually promises. Explain what capability flags, x-mcp-header marks, icons, and cacheScope choices imply about a server's behavior. Spot manifest red flags: a destructive tool with no annotations, a secret mirrored through x-mcp-header, a public cacheScope on user-specific text, and instructions written to steer the model instead of describing it. Parse a registry server.json name into its namespace and state how that namespace was verified. Before a host ever calls a tool, it already knows three things about the server: what server/discover claims to support, what tools/list currently offers, and, if the server is published, what its registry server.json says about who owns it. Nothing in the protocol forces any of those three documents to be complete, cautious, or honest. Annotations are hints. Instructions are self-reported prose. A registry name is only as…
Reading a Server Manifest Like a Reviewer: A server's discover result, its tool list, and its registry entry are the only things you know about it before you…
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.