Deploy a universal-tool-server: You can use the example tool server or create your own.
Launch the ReAct agent locally: Use the tool server URL and API key to launch the ReAct agent locally.
Deploy on LangGraph Cloud: Follow the tutorial here: https://langchain-ai.github.io/langgraph/cloud/quick_start/, but fork this repository instead of the regular react agent template.

Deploy a Universal Tool Server

Using the Example Tool Server

Get the code

git clone git@github.com:langchain-ai/example-tool-server.git

Run the tool server

cd example-tool-server
# Only disable auth if you are running the tool server locally.
# This is done for simplicity during development.
DISABLE_AUTH=true uv run uvicorn app.server:app

See the example-tool-server for more information on how to deploy and configure the tool server.

From Scratch

Install the universal-tool-server package.
```
pip install universal-tool-server
```

Create a new Python file called server.py and implement a simple tool server:

from universal_tool_server import Server

app = Server()

# Define tools
@app.add_tool(permissions=["group1"])
async def echo(msg: str) -> str:
    """Echo a message."""
    return msg + "!"

Run the tool server.
```
uv run uvicorn server:app
```

Launch the ReAct Agent

If you already have a tool server running, you can launch the ReAct agent with the following command:

# Specify at least one chat model API key!
export ANTHROPIC_API_KEY="..."
export OPENAI_API_KEY="..."
# If you enabled auth on the tool server, set the API key.
# If you disabled auth, you set it to a non empty placeholder string.
export TOOL_SERVER_API_KEY="Authorization header token"
# Set the tool server URL
export TOOL_SERVER_URL="http://localhost:8000"
# Launch the ReAct agent
uv run langgraph dev