Files
Alberto Schiabel 684a392816 chore(python): prepare 0.21.1 release (#4361)
## Summary

- bump the Python SDK and all provider package versions to `0.21.1`
- regenerate the root `uv.lock` from the updated workspace metadata
- keep the existing coordinated changelog as the release authority

## Verification

- `pnpm test:release-workflow`
- `make build` (26 artifacts)
- `python -m twine check python/dist/*`
2026-09-04 20:50:30 +02:00
..
2026-07-10 06:14:38 -07:00

composio-langgraph

Adapts Composio tools to LangChain's StructuredTool format for use in LangGraph agents and graph workflows, giving them access to 1000+ apps through a single Composio session.

Installation

pip install composio composio-langgraph langgraph langchain langchain-openai

Set COMPOSIO_API_KEY (get one from dashboard.composio.dev/settings) and OPENAI_API_KEY in your environment:

export COMPOSIO_API_KEY=xxxxxxxxx
export OPENAI_API_KEY=xxxxxxxxx

Quickstart

Create a session for your user, fetch its tools, and hand them to your agent. The wrapped tools also work anywhere LangGraph accepts LangChain tools, such as a ToolNode in a custom graph.

from composio import Composio
from composio_langgraph import LanggraphProvider
from langchain.agents import create_agent
from langchain_openai import ChatOpenAI

composio = Composio(provider=LanggraphProvider())
llm = ChatOpenAI(model="gpt-5.2")

# Each session is scoped to one of your users
session = composio.create(user_id="user_123")
tools = session.tools()

agent = create_agent(tools=tools, model=llm)
result = agent.invoke(
    {
        "messages": [
            (
                "user",
                "Send an email to john@example.com with the subject 'Hello' and body 'Hello from Composio!'",
            )
        ]
    }
)

print(result["messages"][-1].content)

Error handling

Each wrapped tool builds its args_schema from the Composio tool's input schema. When argument validation fails, the tool does not raise; it returns a structured result:

{"successful": False, "error": "<validation message>", "data": None}

Check successful in tool output instead of wrapping calls in try/except.