← back
Tool Calling Is Not Just Plumbing for AI Agents — Roy Derks
Original: Tool Calling Is Not Just Plumbing for AI Agents — Roy Derks
Takeaway
Agents are only as good as their tools — invest in well-described, schema-typed, framework-independent tool layers, not just the agent loop.
Summary
- Argues tool quality, not agent framework, determines agent quality; tools should be reusable, robust, framework-portable.
- Tool definitions deserve system-prompt-level care: name, long description, input params, plus increasingly important output schemas for chained/structured calls.
- Distinguishes traditional tool calling (logic in agent server) vs Cloudflare-style remote tools and meta-tool-calling that lets the model generate SQL/GraphQL dynamically at runtime.
- Dynamic tools backed by GraphQL/SQL let one tool surface entire APIs/databases without enumerating every endpoint.
- Encourages separating tool platform from agent framework so swapping frameworks doesn't break tools.
tool-callingagentsmcp
Original description
Tool calling is more than just a technical detail—it’s the most important piece of how AI agents get things done. It allows agents to connect to systems, access data, and perform all sorts of tasks. Building good tools is very similar to building good APIs, as they need good design for things like authentication and security. In this session, we’ll explore how tool calling works, why it matters, and how to do it right. You’ll learn different ways to build tools that help agents connect to APIs, databases, and other resources, making them smarter and more useful. Build a MCP server: https://hackteam.io/blog/build-your-first-mcp-server-with-typescript-in-under-10-minutes/ Tool Calling platform: https://github.com/IBM/wxflows