What is tool metadata?
Tool metadata is the tool-definition information an AI application receives, such as a name, a functional description, an input parameter schema and, optionally, an output schema and behavior annotations. This information describes how a tool is meant to be used; it is neither the tool’s implementation nor proof that a call succeeded.
What each field answers
The name identifies the tool. The tool description explains what the tool does, what it is for and when to use it. The input schema defines the parameters’ fields, types and required items. An output schema can define the format of structured results. The fields supported can differ across protocols and clients.
Take the MCP 2025-11-25 specification as an example: a tool definition includes name, description and inputSchema, and can also provide outputSchema and annotations. Annotations describe behavior; they cannot replace the server’s permission checks.
Which fields are visible depends on the application
A tool catalog that holds the full definition does not mean the model receives all of it on every turn. Some hosts run candidate retrieval first and then load only some of the tools; the fields used for search indexing and the fields the model finally sees can also differ.
TSO looks at how a tool is found and selected; TCC goes a step further and checks the task, the arguments and how the return value is interpreted. Even clearly written metadata still needs real logs to verify behavior.
Sources and scope
Sources as checked for the source-language entry on 2026-11-06; this English edition adds no new claims and does not represent a new check.