Tool Use Pattern: Designing Agent Tool Sets
The Tool Use pattern is the mechanism by which agents interact with external systems. How you design the tool set — names, descriptions, scope, and response format — determines whether the agent is reliable or unpredictable.
What Is the Tool Use Pattern?
The Tool Use pattern describes how agents interact with external systems through tools — and more specifically, how to design those tools so the agent uses them reliably. An agent with a well-designed tool set behaves predictably. An agent with a poorly designed tool set guesses constantly and fails unpredictably.
When to Use It
Every agent that takes action in enterprise systems uses the Tool Use pattern. It's not a pattern you choose — it's a pattern you implement well or poorly. The question is how to design your tool set for maximum agent reliability.
Core Tool Design Principles
- Single purpose per tool — each tool does one thing. Overloaded tools that handle multiple actions through parameter combinations are the leading cause of agent unreliability.
- Business-action naming — tool names describe business actions, not technical operations.
approve_purchase_order, notupdate_po_status. - Complete descriptions — the description tells the LLM when to call the tool, what it needs as input, and what it will return. Incomplete descriptions cause wrong tool selection.
- Structured responses — every tool returns a consistent, LLM-readable response format. The agent reasons based on what you return.
Tool Set Size and Scope
Keep tool sets between 10 and 25 tools per MCP server. Below 10, you may be missing important capabilities. Above 25, the LLM begins to have difficulty distinguishing between similar tools. Use the Orchestration-Federated pattern to give agents access to more tools while keeping individual server sets manageable.