A tool owns one callable capability. It defines input, output, and execution.
Tools give agents and workflows defined ways to act. The model can choose a tool, but the tool owns the code that runs.
Characteristics
- Typed input describes what the caller must provide.
- Execution performs one operation.
- Output returns a structured result the caller can use.
- Errors describe why the operation could not complete.
Boundary
Tools can be local project files, remote integration tools, or MCP-exposed capabilities. The caller chooses when to invoke them. The tool owns how the work runs.
Keep tools focused. A tool should do one thing, validate its input, and return a clear result. If the operation grows into multiple stages, approvals, or retries, move the coordination into a workflow.
Wrong fit
Do not use a tool as a hidden workflow, long-running run, or large integration layer. Use a workflow for process, a run for durable execution, and an integration for reusable external service access.
For implementation steps, see Tools.