What is an agent tool?
A typed capability that an AI agent invokes to retrieve data, change state, execute code, call an API, or interact with an external environment.
Why is this important?
Tools give agents grounded information and real effects. Their schemas and permissions determine what the agent can safely do.
How it works
A tool exposes a name, description, input schema, execution handler, and optional output schema. The runtime validates arguments, executes, and returns a structured result.
Technical example
A create-task tool validates customerId, title, ownerId, and dueAt, writes idempotently, and returns the created task ID.
Implementation notes
Prefer narrow tools with strict schemas, bounded output, deterministic errors, idempotency keys, least privilege, audit logs, timeouts, and clear read versus write semantics.
Sources
https://modelcontextprotocol.io/docs/learn/server-conceptshttps://modelcontextprotocol.io/docs/learn/architecture
Watch a video explanation
