navra

navra + LangGraph

Prerequisites

  • navra running with upstream MCP servers configured
  • A bearer token
  • Python with langgraph, langchain-openai, and openai installed

Architecture

LangGraph agents use LLM tool calling to interact with tools. navra exposes tools through its OpenAI-compatible /v1/chat/completions endpoint, so LangGraph agents can call navra-proxied MCP tools through the standard OpenAI function calling protocol.

LangGraph Agent → OpenAI SDK → navra /v1 → MCP upstream tools
                                    ↓
                              Safety filters
                              ACL enforcement
                              Audit logging

Basic setup

from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

llm = ChatOpenAI(
    base_url="http://localhost:9315/v1",
    api_key="mcd_your_token_here",
    model="granite3.3:8b",
)

agent = create_react_agent(llm, tools=[])
result = agent.invoke({
    "messages": [{"role": "user", "content": "List files in /tmp"}]
})

Using navra's MCP tools

navra automatically exposes upstream MCP tools through its OpenAI endpoint. The model sees them as function calls. No manual tool registration is needed -- navra's tool discovery handles it.

If you want to restrict which tools LangGraph can use, configure ACLs in navra's permission set:

[permissions.langgraph]
safety = "standard"
ring = 2
allow = ["file_read", "file_list"]
deny = ["file_write", "file_delete"]
operations = ["read"]

With custom LangChain tools alongside navra

You can mix navra-proxied tools with native LangChain tools:

from langchain_core.tools import tool

@tool
def calculate(expression: str) -> str:
    """Evaluate a math expression."""
    return str(eval(expression))

agent = create_react_agent(
    llm,
    tools=[calculate],  # local tools
)

The local tools run in-process. navra-proxied tools run through the gateway with full security enforcement.

Streaming

for chunk in agent.stream({
    "messages": [{"role": "user", "content": "Analyze README.md"}]
}):
    print(chunk)

Troubleshooting

LangGraph cannot reach navra

Verify the base URL and token:

from openai import OpenAI
client = OpenAI(base_url="http://localhost:9315/v1", api_key="mcd_...")
print(client.models.list())  # should return available models

Tool calls are not routed through navra

Make sure you are using ChatOpenAI with navra's base URL, not a direct OpenAI or Ollama connection. Tools defined as @tool decorators in LangChain run locally and bypass navra.

Rate limiting

If running many agent iterations, navra may throttle requests. Check navra status for rate limit counters.