Warum Schemas besser sind als lose Argumente

When a tool takes a plain dict or untyped values, the model can send almost anything, and your code only finds out when it crashes. 😬

A schema is a clear contract: it states exactly which fields exist, their types, and which are required before any code runs.

With a schema, bad input is rejected at the boundary, so your tool body always works with data you can trust.

That same schema is shown to the model, so the AI learns the exact shape it must produce instead of guessing field names.

Here a tool just grabs whatever keys it hopes are present, with no guarantee they actually exist or hold the right type.

Without a schema you scatter manual if checks everywhere. A validated input lets you delete that boilerplate entirely.

A schema turns a vague crash into a precise error: it names the bad field and says what was expected.

Pydantic is the library that powers this in MCP: you describe input as a model and it validates and documents it for you.

The model definition is the single source of truth: the JSON schema, validation, and docs all flow from that one class.

Because the schema lists every field and type, a typed tool is self-documenting, helping both new humans and the model.

As your server grows, schemas keep tools predictable, so dozens of tools behave consistently instead of each parsing input its own way.

Why prefer a schema over accepting a loose dict in a tool?

A schema is a contract that validates input early, documents the shape for the model, and replaces scattered manual checks. ✅