텍스트와 구조화된 콘텐츠

Every MCP tool result carries content, and you choose its shape. The big split is plain text versus structured data the client can parse.

A text result is just a string for the model to read, like a sentence or a paragraph. It is simple, human-friendly, and easy to drop into a prompt. 📝

Structured content is real data, like a JSON object with named fields. Clients can read each field instead of guessing from prose.

If a tool returns "It is 21 degrees and sunny," the client must parse that English to use it. Wording changes break the parsing instantly.

With structured output, a client reads result.temp_c directly. The shape is predictable, so apps and agents can rely on it without scraping text.

MCP lets a tool return a readable text summary and structured data together. The model reads the prose; code reads the fields.

Pick plain text when the answer is a message for a human to read, like an explanation, a status note, or a short summary.

Pick structured content when another program will consume the result, such as a price, a record, or a list of items to act on.

If your tool just returns a string, MCP wraps it as text content automatically. Structured output is something you opt into on purpose.

Ask who reads the result: a human, the model, or downstream code? That consumer decides whether you reach for text or structure.

Mature tools usually return both: a friendly summary for the model plus a typed payload for the app. This keeps the result useful on every side.

A tool feeds prices into a billing app. Which output fits best?

You learned the two result shapes: readable text for humans and the model, and structured data for code. Match the shape to who reads it. ✅