Tool Calling
The mechanism by which a language model invokes external functions — APIs, databases, code execution, retrieval — and reads the results back to continue its work.
Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.
What it is
Tool calling (also called function calling) is how a language model interacts with the world beyond text generation. The model is told which tools it can use, with what arguments. When it decides to call one, it emits a structured request; the runtime dispatches the actual call; the result is fed back into the model's context. The model can call multiple tools across a single task.
Why it matters
Tool calling is what turns a language model from a text generator into an agent that can do work. Without tools, the model can only describe actions; with tools, it can take them — read files, call APIs, run code, update records. Every meaningful agent capability passes through tool calling.
How it works
Tools are declared with typed schemas (JSON Schema is the de facto standard). The model receives the schemas as part of its context. When it generates a call, the runtime validates the arguments, dispatches the call, and feeds the response back. Modern best practice exposes tools through the MCP Tool Registry so governance, audit, and permission scopes apply uniformly.
Related resources
A governed catalog of every tool an AI agent can call — your APIs, your databases, your internal systems — with typed schemas, permission scopes, audit trails, and the standard protocol (MCP) that turns 'we exposed it to the LLM' into 'we know exactly who called what when'.
Anthropic's open standard for exposing tools, resources, and prompts to AI models — released in late 2024, broadly adopted across the agent ecosystem, the connective tissue of modern AI tool integration.
Each AI agent gets its own limited set of keys — it can only use the tools its job requires, with credentials that are scoped, temporary, and tracked.
Every tool an agent uses comes with a precise, versioned form describing exactly what goes in and what comes out — so nothing changes shape quietly and breaks your workflows.
A complete, tamper-proof record of every action your agents take — which tool, on whose behalf, with what inputs and what result — so any question about what happened has an answer.