Skip to content

Composio

SDK for building AI agents that can take actions in 1000+ apps via pre-authenticated tools.

Stars
30.9k
Forks
4.9k
Licence
MIT
Language
TypeScript
Last update
4 days ago
Type
Dev tool

What it is

Composio is a TypeScript and Python SDK, plus a command-line tool, that lets AI agents use external apps on a user's behalf. It provides pre-authenticated toolkits for over 1000 apps, per-user sessions, authentication, triggers, and a sandbox. Developers use it to connect an agent framework to real services such as email.

  • Create a per-user session and give an agent tools to summarize that user's emails
  • Expose a session's tools to an MCP client such as Claude or Cursor via a hosted endpoint
  • Search, execute, and connect app tools from the shell with the composio CLI
  • Script multi-step tool workflows in TypeScript with composio run

Great for

  • Developers building AI agents that need to read or act in third-party apps
  • Teams already using a supported agent framework such as OpenAI Agents, LangChain, or Vercel AI SDK
  • Projects that want per-user scoping of app connections
  • Developers who want to expose tools to MCP clients like Claude or Cursor

Not ideal for

  • Non-developers looking for a finished end-user app, since the README describes an SDK and CLI
  • People who need a fully self-hosted setup, because the README's quickstart relies on a Composio API key and dashboard
  • Projects needing a framework the README does not list as supported, which would require writing a custom provider
Step-by-step guide: how to use ComposioInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Install with a package manager

    You'll need: Node 22+ for the TypeScript SDK · Python 3.10+ for the Python SDK · A COMPOSIO_API_KEY from the Composio dashboard

    Install TypeScript SDK with OpenAI Agents provider

    npm install @composio/core @composio/openai-agents @openai/agents

    Install Python SDK with OpenAI Agents provider

    pip install composio composio-openai-agents openai-agents

    Install the CLI

    curl -fsSL https://composio.dev/install | sh
  2. 2

    Run it

    • Get a COMPOSIO_API_KEY from the dashboard settings page
    • Install the SDK and a provider package for your framework
    • Create a session for a user with composio.create("user_123") and call session.tools()
    • Pass the tools to your agent and run it with a request such as summarizing today's emails
  3. 3

    Get the most out of it

    • Store session.session_id and reuse it with composio.use() across turns
    • Restrict toolkits, auth configs, and connected accounts using the configuring sessions docs
    • Pass mcp: true to composio.create() to get a hosted MCP endpoint for MCP clients
    • Use @composio/slim for a smaller install with the same API, per the README

Downloads

Strengths

  • Broad coverage: the README lists 1000+ toolkits and many provider adapters for common agent frameworks
  • Sessions load meta tools at runtime instead of hundreds of tool definitions, per the README
  • Available in both TypeScript and Python with an MIT license

Trade-offs

  • Depends on a Composio account and API key for the quickstart, so you rely on a hosted service
  • The README is truncated and gives limited detail on sandbox, triggers, and pricing
  • Some providers are experimental, such as the Pi provider in @composio/experimental

How it compares

ProjectStarsTypeLevelUpdatedBest for
composio (this)30.9kDev toolintermediate4 days agoDevelopers building AI agents that need to read or act in third-party apps
langchain149kFrameworkintermediate4 days agoPython developers building LLM-powered applications
llama_index52.8kLibraryintermediate10 days agoPython developers building retrieval-augmented generation (RAG) apps over private documents
crewAI60.4kLibraryintermediate4 days agoPython developers prototyping multi-agent workflows
SuperAGI17.7kAI model / agentadvanced1.8 years agoPython developers experimenting with autonomous AI agents
plandex15.9kCLIintermediate1.1 years agoDevelopers working in large codebases who want an AI agent to handle multi-file changes

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.