MCP y llamadas a herramientas

Composio

composiohq/composio

Composio is an open-source SDK monorepo for connecting AI agents to more than 1,000 toolkits. It provides tool discovery and execution, per-user sessions, authentication, triggers, context management, hosted MCP endpoints, provider adapters, a CLI, and a sandboxed workbench.

Abrir repositorio oficial ↗ Página de inicio del proyecto ↗
★ 29,5KEstrellas
⑂ 4,7KBifurcaciones
98Problemas abiertos
TypeScriptIdioma
MITLicencia
Q@project.QualityScorePuntuación editorial

Resumen

The repository contains TypeScript and Python SDKs, a command-line interface, provider adapters, experimental integrations, and documentation. Applications create a user-scoped session, expose its tools to an agent framework, and let the agent discover, authenticate, and execute app actions. Sessions can also be presented to compatible clients through hosted MCP endpoints.

Características principales

  • More than 1,000 agent toolkits
  • Per-user sessions with reusable session IDs
  • Runtime tool discovery, authentication, and execution through meta tools
  • Hosted MCP endpoint for each MCP-enabled session
  • TypeScript and Python SDKs
  • CLI for searching, executing, linking, and scripting tools
  • Provider adapters for multiple agent frameworks and model SDKs
  • Authentication, triggers, context management, and a sandboxed workbench
  • Configurable toolkit, authentication configuration, and connected-account restrictions
  • Slim TypeScript package with the same API but without packaged source and SDK documentation

Requisitos, instalación y guía rápida

TypeScript quick start: `npm install @composio/core @composio/openai-agents @openai/agents`. Python quick start: `pip install composio composio-openai-agents openai-agents`. The documented CLI installation is `curl -fsSL https://composio.dev/install | bash`, followed by `composio login`. For repository development, run `mise install`, `pnpm install`, `pnpm build`, and `pnpm test`; Python commands run from the `python/` directory.

Uso

Obtain a COMPOSIO_API_KEY from https://dashboard.composio.dev/settings. Initialize `Composio` with the provider adapter required by the selected framework, create a session for a user identifier, retrieve `session.tools()`, and supply those tools to the agent. Preserve `session.session_id` and restore it with `composio.use()` across turns. For MCP, pass `mcp: true` when creating the session and configure the MCP client with `session.mcp.url`. CLI users can use `composio search`, `composio execute`, `composio link`, and `composio run`.

Compatibilidad de modelos y casos de uso

The repository lists adapters for OpenAI, OpenAI Agents, Anthropic, Claude Agent SDK, Vercel AI SDK, Google GenAI, Google ADK, LangChain, LangGraph, LlamaIndex, Mastra, Pi, Cloudflare Workers AI, CrewAI, and AutoGen, with language availability varying between TypeScript and Python. LangGraph is available through the TypeScript LangChain adapter and through a dedicated Python package. The Pi adapter is explicitly experimental. Custom providers and MCP are offered for frameworks not covered by an adapter. Exact supported model versions are not stated in the repository metadata.

Notas sobre la licencia y los riesgos

The repository declares the MIT license and links to https://api.github.com/licenses/mit. The README also references the repository LICENSE file. No additional license exceptions or dependency-license details are stated in the supplied record.

Lanzamiento y mantenimiento

The default branch is `next`. The supplied metadata reports that the repository was created on 2024-02-23 and last pushed on 2026-08-03. Specific release versions and release contents are not stated in the repository metadata. The project changelog is available at https://docs.composio.dev/reference/changelog.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython