What is an MCP Tool?

MCP tools are the action primitives of the protocol. They represent things an agent can do — execute a shell command, query a database, create a file, send an API request. Each tool has a schema describing its inputs and expected outputs.

What makes MCP tools powerful is standardisation. A tool defined once on an MCP server is automatically available to any compatible agent. The agent's LLM sees the tool's name, description, and parameters — enough to decide when and how to use it.

Tool design matters enormously. Well-designed tools have clear names, comprehensive descriptions (the LLM reads these to decide when to use the tool), strict input validation, and informative error messages. But even well-designed tools can be misused — which is why tool-level policy enforcement matters.

PolicyLayer enforces policies at the individual tool level. Each grant's policy specifies which tools are allowed or denied, with optional require and deny_if conditions on arguments. For example, you can allow the execute_sql tool but deny DROP statements, or allow write_file only to specific directories. Each tool call is evaluated against the policy before reaching the server.