Tool Type Generator - Composio

View source on GitHub

This cookbook shows how to access the raw tool definitions and toolkit metadata that Composio exposes. If you are building a platform on top of Composio, say a workflow builder or an IDE plugin, you will need to display tool schemas, auth requirements, and input/output parameters to your users. This is how you get that data.

Prerequisites

Python 3.10+
UV
Composio API key

Project setup

Create a new project and install dependencies:

bash

mkdir composio-tool-generator && cd composio-tool-generator
uv init && uv add composio

Add your API key to a .env file:

bash

COMPOSIO_API_KEY=your_composio_api_key

Setting up the client

No provider is needed here. We are reading tool metadata directly from the Composio API, not passing tools to an AI framework.

<include>../../examples/tool-generator/main.py#setup</include>

Raw tool definitions

Composio has two internal representations for tools: raw definitions and provider definitions. Raw definitions are framework-agnostic JSON schemas with input_parameters, output_parameters, scopes, and tags. Provider definitions translate these into the format a specific framework expects (OpenAI, Anthropic, etc.).

For building platform UIs, the raw definitions are what you want. composio.tools.get_raw_composio_tools() returns them for an entire toolkit. Each tool has a standard JSON Schema for its inputs and outputs, so you can render forms, validate user input, or generate types from it.

<include>../../examples/tool-generator/main.py#list-tools</include>

Toolkit metadata

A toolkit groups related tools under one integration. composio.toolkits.get() returns metadata about the toolkit itself, including its supported auth schemes. Each auth scheme lists the fields required for two stages: creating an auth config (platform setup) and initiating a connected account (end-user connection). This is useful if you need to build your own auth configuration UI.

<include>../../examples/tool-generator/main.py#toolkit-info</include>

Exporting to JSON

For code generation or static analysis, you can export the full raw schemas to a JSON file. Each entry includes the input_parameters and output_parameters as standard JSON Schema objects, which you can feed into tools like json-schema-to-typescript or datamodel-code-generator to produce typed interfaces.

<include>../../examples/tool-generator/main.py#export</include>

Complete script

Here is everything together:

<include>../../examples/tool-generator/main.py</include>

Running the script

List all tools in a toolkit:

bash

uv run --env-file .env python main.py list-tools gmail

Inspect a toolkit's auth schemes:

bash

uv run --env-file .env python main.py toolkit-info gmail

Export raw tool schemas to a JSON file:

bash

uv run --env-file .env python main.py export gmail tools.json

Take it further

Raw tool schemas are building blocks for platform features:

Dynamic form builder: use input_parameters JSON Schema to render forms for any tool — no hardcoded UI per integration
Type generation pipeline: feed exported JSON into json-schema-to-typescript or datamodel-code-generator in CI to keep types in sync
Tool catalog UI: combine list-tools and toolkit-info to build a searchable directory of available integrations and their auth requirements