MCP Tool Schema

MCP folds tool capabilities into a unified input schema. When you write an Agent, you don't have to type a full JSON by hand.

Fill in Tool Name / Description / Parameters, or describe it in natural language, then click Generate MCP Tool.

The output is a Schema assembled by local rules — treat it as a draft. Real enums, defaults, and auth fields still follow the service.

It pairs with the Function Calling page: that one is closer to OpenAI tools, this one to an MCP tool list.

Write Tool Name, Description, and a parameter list separately. One parameter per line, mark required ones — that infers better than a paragraph.

Copy name, description, and inputSchema into your server. This page does not invoke tools and does not connect to any MCP host.

How to use it

Fill in a tool name, description, and parameters, or write a natural-language paragraph.
Click Generate MCP Tool and check that parameter types look right.
Add enum, default, and required. Remove parameters that should not be exposed.
Copy the Schema into your MCP server list.
If you also need an OpenAI tools wrapper, export the same parameters on the Function Calling page.

Register tools such as search_repo and read_file for an IDE Agent.
Draft MCP tools from an internal HTTP API.
Share fields with Structured Output, but the purpose here is tool input.
Review whether the tool surface is too wide: one tool should not have twenty optional parameters.

You don't connect to an MCP server, and the description is not sent.
Parameter types are guessed from names. Edit numbers and enums by hand.
Use a stable snake_case tool name so old clients don't break when you rename later.

How is this different from Function Calling?

MCP is aimed at a tool list and input_schema. Function Calling is closer to OpenAI / Gemini tools wrappers.

Are parameter types accurate?

Common types are inferred from names and descriptions. Numbers, enums, and required still need to match the real API.

Do I need to connect to an MCP server?

No. This page only generates a definition. It does not invoke a tool.

Is the description uploaded?

No.

Should I write an output schema?

This page focuses on input. Keep output shapes on Structured Output or the Schema page.

Can it generate resources / prompts?

This page only covers Tool. Write Resource and Prompt by the MCP spec.

How does this map to OpenAPI?

One HTTP operation often becomes one MCP tool. Path params and query become input fields.

What if names collide?

name must be unique on the same server. Search the list after generate.