A tool call is a request issued by an AI agent to invoke a specific function or capability exposed through the Model Context Protocol (MCP). When an agent determines that it needs to perform an action—such as querying a database, calling an API, reading a file, or executing code—it formats a tool call request with the appropriate function name and parameters. The MCP server receives this request, validates it, executes the corresponding function, and returns the results back to the agent.

Tool calls are fundamental to agentic workflows because they enable agents to interact with systems beyond their training data. Rather than hallucinating answers, agents can invoke real tools to fetch current information, perform calculations, or trigger actions. This creates a feedback loop where the agent's reasoning is grounded in actual system responses. Tool calls require clear schema definitions that specify function names, required parameters, parameter types, and expected return values, allowing both the agent and the MCP server to communicate effectively.

In practical AI applications, tool calls enable multi-step problem-solving where agents decompose complex tasks into sequences of function invocations. For example, an autonomous coding agent might call tools to read repository files, analyze code, run tests, and commit changes—each call producing concrete results that inform the next step. The quality and reliability of tool calls directly impacts agent effectiveness, making proper error handling and timeout management critical considerations in production systems.