Crabhaus

What it is

The API mechanism that lets a language model use external tools. You send the model a list of tool definitions (each one a name, a description, and a JSON Schema for its parameters) alongside the conversation. Instead of answering in prose, the model may reply with a structured request to call one of those tools.

How it works

A request/response loop in which the model never executes anything. Each API request carries the tool schemas plus the conversation. When a tool would help, the model emits the chosen tool's name and arguments as JSON. Your code, not the model, runs the real function, appends the result as a tool-result message, and calls the API again; the model reads it and answers or asks for another call.

A concrete example

Ask 'what's the weather in Dublin?' with a get_weather tool defined. The model returns a call like {"name": "get_weather", "arguments": {"city": "Dublin"}}. Your harness hits a real weather API, sends back {"temp_c": 17, "raining": true} as the tool result, and a second model turn produces 'It's 17C and raining in Dublin.'

Why it matters

Agent frameworks, MCP servers, and coding assistants are all this one loop with more tools and retry logic on top. Knowing it explains agent costs (every tool's schema rides along on every turn), latency (each call is a full round trip), and reliability (arguments are model-written and can be wrong even when schema-valid).