langchain_2ly

2ly python module providing helpers to quickly connect 2ly to your agents in Python.

Usage

Install the module

pip install langchain_2ly

Connects your LangGraph agent:

# Import MCPAdapter
from langchain_2ly import MCPAdapter
mcp = MCPAdapter("my-runtime")
tools = await mcp.get_langchain_tools()
agent = create_react_agent(llm, tools)
await mcp.stop()

Under the hood our connector leverages Langchain MCP adapters. You can also use our variant which is strictly based on the Official MCP Python SDK (see Using MCP without Langchain MCP Adapter).

MCPAdapter lifecycle

MCP Adapter will start an MCP Server lazily when you first call get_langchain_tools() or when the agent call a given tool. In practice the call to retrieve the list of tools must always come first.

Since this is an async task, we provide a way to instantiate MCPAdapter with async and thus it will automatically stop itself when the agent is done.

async def main():
    async with MCPAdapter("My agent name") as mcp:
        tools = await mcp.get_langchain_tools()
        agent = create_react_agent(llm, tools)
        agent_response = await agent.ainvoke() # as usual

if __name__ == "__main__":
    asyncio.run(main())

Using MCP without Langchain MCP Adapter

Simply replace MCPAdapter with MCPClient. The API is identical.

```python
async def main():
    async with MCPClient("My agent name") as mcp:
        tools = await mcp.get_langchain_tools()
        agent = create_react_agent(llm, tools)
        agent_response = await agent.ainvoke() # as usual

if __name__ == "__main__":
    asyncio.run(main())

Usage examples

• examples/list_tools.py: list tools available for a given 2ly agent
• examples/langgraph.py: connect 2ly to a LangGraph agent
• examples/langgraph_without_adapter.py: another LangGraph example but without Langchain adapter

Options

Both variants accept optional configuration to control how the MCP runtime is started via npx and how the client behaves:

from langchain_2ly import MCPAdapter, MCPClient

options = {
  "workspace": "my-workspace-id",           # forwarded to runtime as WORKSPACE_ID
  "nats_servers": "nats://localhost:4222",  # NATS URL used by runtime
  "version": "latest",                      # @2ly/runtime version for npx
  "startup_timeout_seconds": 20.0,            # wait for initialize()
  "log_level": "info",                       # forwarded to runtime as LOG_LEVEL
}

mcp = MCPAdapter("my-runtime", options)
# or
mcp = MCPClient("my-runtime", options)

Development

Prepare your venv

cd packages/langchain_2ly
python3.11 -m venv .venv # any version python3.10+ will do
source .venv/bin/activate
pip install --upgrade pip
pip install -e ".[all]"

Execute tests

pytest

Build locally

python -m build

Test local installation

# update the filename to the build version
# update the filename to the build version
pip install dist/langchain_2ly-0.0.4-py3-none-any.whl --force-reinstall

Run the examples

python examples/list_tools.py
python examples/langgraph_agent.py
python examples/langgraph_without_adapter.py