Tools Primitive: Model-Controlled Actions with Side Effects

~15 min read

Tools are executable functions the AI can invoke, usually triggered by the model's own choice — but since they can have real side effects, MCP implementations often require explicit user permission before execution.

Tools Primitive: Model-Controlled Actions with Side Effects is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Tools are executable functions the AI can invoke, typically operations with real effects or computation beyond the AI's own capabilities
  • Usually model-controlled: the LLM itself decides when it needs a tool's functionality
  • Registered on the server (e.g. via @mcp.tool()), invoked via tools/call with a name and arguments, returning structured results
  • Since tools can have side effects (file I/O, network calls), MCP implementations often require explicit user permission before execution
  • Analogous to classic function-calling 'functions,' but used in MCP's more flexible, dynamic, discoverable context