# LangGraph

URL: https://docs.hanzo.ai/docs/guides/integrations/langgraph

> The base URL goes on the model, not the agent. ChatOpenAI takes base_url; hand the instance to create_agent.

The base URL goes on the model object, not on the agent. `ChatOpenAI` takes **`base\_url`** (alias `openai\_api\_base`), and `create\_agent`accepts an initialized model instance — which is the path that carries a custom base URL, since the`"provider:model"\` string form cannot.

```bash
pip install langchain-openai langgraph
```

```python
from langchain_openai import ChatOpenAI
from langchain.agents import create_agent

model = ChatOpenAI(
    model="zen6",
    base_url="https://api.hanzo.ai/v1",
    api_key="sk-...",
)

agent = create_agent(model=model, tools=[])
result = agent.invoke({"messages": [{"role": "user", "content": "Hello"}]})
```

Lands on `POST /v1/chat/completions`.

Resolution order for the address, first match winning: the explicit `base_url` or `openai_api_base` argument, then `OPENAI_API_BASE` which LangChain reads at init, then `OPENAI_BASE_URL` which the underlying `openai` client reads. Setting the argument makes the other two irrelevant, which is why the snippet sets it.

## MCP

`langchain-mcp-adapters` holds the client. The URL is the `url` key of a per-server entry.

```bash
pip install langchain-mcp-adapters
```

```python
from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "hanzo": {
        "transport": "http",
        "url": "https://api.hanzo.ai/v1/mcp",
        "headers": {"Authorization": "Bearer sk-..."},
    }
})

tools = await client.get_tools()
agent = create_agent(model=model, tools=tools)
```