What is an MCP Tool

An MCP tool is an executable function that an AI model can discover and invoke through the Model Context Protocol. Tools let AI take actions in the real world — querying databases, creating files, calling APIs, running shell commands.

How it works

Each tool is defined by a name, a description, and an input schema:

{
  "name": "search_files",
  "description": "Search for files matching a pattern in the project",
  "inputSchema": {
    "type": "object",
    "properties": {
      "pattern": { "type": "string" },
      "path": { "type": "string" }
    },
    "required": ["pattern"]
  }
}

The AI reads the description and schema, decides when the tool is relevant, constructs the input arguments, and sends a tools/call request. The MCP server executes the function and returns the result. The AI observes the result and decides what to do next.

Tools are model-controlled — the AI decides when and how to use them. This is what separates tools from resources, which are application-controlled.

Why it matters

Without tools, AI can only generate text. It can describe what you should do but cannot do it. Tools close this gap. An AI with access to a run_query tool can answer "what were last month's sales?" by actually running the query, not guessing from training data.

The tool abstraction also creates composability. A single MCP server can expose dozens of tools. An AI agent can connect to multiple servers and combine tools from different domains in a single workflow — search files, read documentation, write code, run tests.

See How MCP Tools Work for the full lifecycle from discovery through execution.