MCP and AI basics
Tool schema
The machine-readable description of a tool: its name, what it does and the inputs it accepts, written in JSON Schema. The model reads it to know how to call the tool correctly.
A tool schema is the machine-readable description of one tool: its name, a plain-text description of what it does and a JSON Schema for the inputs it accepts. The model reads the schema to decide whether the tool fits and to construct a valid call. A GA4 report tool, for example, might declare a property ID, a date range, a list of metrics and a list of dimensions.
Schemas shape how well an assistant uses a server. Clear names and descriptions lead to correct calls; vague ones lead to guessed parameters and odd results. When a server feels unreliable, reading its tool list in a developer tool such as the MCP Inspector often explains why.
Schemas are also a security surface. The description text is read by the model, so a malicious server can hide instructions in it. This is called tool poisoning, and it is one reason to prefer vendor-published servers and to look at what a community server actually exposes before connecting a real account.