Glossary
function calling
A pattern where a model emits a structured call (function name plus arguments), the runtime executes it, and the result returns as input on the model's next turn.
The protocol that lets a language model trigger code execution. The caller defines a tool schema (name, description, JSON-schema parameter shape); the model either responds with text or emits a structured tool- call object naming the tool and arguments (a form of structured outputruntimeMaking a model emit text that conforms to a schema or type (often JSON matching a JSON Schema) rather than free-form prose, so downstream code can parse it reliably. Open full entry ); the host runs the tool and returns the result; the model continues with that result in context. That emit-execute-feedback cycle is the tool-call loopagentsThe host-orchestrated cycle behind tool use: the model emits a tool call, the host (not the model) executes it, and the result is fed back for the next turn. Open full entry .
The technique is what makes agents possible. Web search, code execution, database queries, file reads, MCPprotocolsAn open protocol from Anthropic that standardizes how language models discover and call external tools, data sources, and prompts via a small JSON-RPC interface. Open full entry server actions: all are function calls under the hood. Reliability varies by model: function-calling quality is now a standard axis on which models are benchmarked.
Two flavors coexist. Provider-specific schemas (OpenAI’s tools API,
Anthropic’s tool_use blocks, Gemini’s function declarations) are the
common case. The MCPprotocolsAn open protocol from Anthropic that standardizes how language models discover and call external tools, data sources, and prompts via a small JSON-RPC interface.
Open full entry (MCP) tries to standardize
the tool layer across providers so the same MCP server works regardless
of which model is making the call.