Connect LangChain to Odoo
via MCP

Why connect LangChain to Odoo?

LangChain is where you build AI agents; Odoo is where your business data lives. If you want an agent that can answer questions about real sales orders, open invoices, stock levels, or CRM pipeline — instead of hallucinating — it needs live access to the ERP. This guide shows how to connect LangChain to Odoo through the Model Context Protocol (MCP), so a LangChain Odoo agent can read and query your data directly.

The question "how to connect Odoo to LangChain" usually ends in a pile of custom XML-RPC glue. MCP removes that. LangChain supports MCP tools natively via the langchain-mcp-adapters package, so once Odoo exposes an MCP endpoint, your agent discovers the available tools automatically — the same open standard Claude, Cursor, and other clients already speak.

The result is a clean Odoo LangChain integration: your agent gets structured, governed access to live Odoo records, and you skip building and maintaining a bespoke API layer.

The approach: LangChain consumes Odoo MCP tools

MCP is a standard way to expose tools to AI agents. An Odoo MCP endpoint publishes tools such as search_records, get_record, and describe_model. On the LangChain side, langchain-mcp-adapters' MultiServerMCPClient connects to that endpoint and turns those into native LangChain tools your agent can call.

You just need an Odoo MCP endpoint to point at. KSROlabs provides two, depending on how your Odoo is hosted — a self-hosted native module and a purpose-built gateway for Odoo Online. Both expose the same LangChain Odoo MCP tools, so your agent code is identical either way.

Step-by-step: build a LangChain Odoo agent

Follow these 5 steps to connect LangChain to your live Odoo instance via MCP.

Putting it together, a minimal LangChain Odoo agent looks like this:

from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

# 1. Point LangChain at your Odoo MCP endpoint
client = MultiServerMCPClient({
    "odoo": {
        "transport": "streamable_http",
        "url": "https://your-odoo.com/mcp/v1",
        "headers": {"Authorization": "Bearer <your-scoped-token>"},
    }
})

# 2. Discover Odoo tools exposed over MCP
tools = await client.get_tools()

# 3. Build a LangChain agent that can use them
agent = create_react_agent(ChatOpenAI(model="gpt-4o"), tools)

# 4. Ask a question over live Odoo data
result = await agent.ainvoke({
    "messages": "List our five most recent confirmed sale orders."
})
print(result["messages"][-1].content)

Swap the model for any LangChain-supported LLM, and swap the endpoint URL for your MCP Server or Odoo Online MCP Gateway address. The agent handles tool selection on its own.

Two hosting paths for your Odoo MCP endpoint

How you expose Odoo over MCP depends on how Odoo is hosted. LangChain connects to either the same way:

Read more on each path: the native Odoo MCP Server module for self-hosted and on-prem Odoo, or the Odoo Online MCP Gateway for *.odoo.com SaaS tenants.

What you can build with a LangChain Odoo agent

Once your agent has live Odoo tools, you can build agentic workflows over ERP data:

Keeping the agent's access governed

An autonomous agent with ERP access needs guardrails. The Odoo MCP endpoint is designed for this: every connection uses a scoped API key, keys are read-only by default, and write access is granted per model only when you explicitly enable it. For a LangChain agent, give it a dedicated least-privilege key limited to the models it actually needs.

Bearer tokens can be rotated or revoked at any time, and requests are logged so you keep an audit trail of what the agent read or changed. That means you can adopt an Odoo LangChain integration without handing an LLM unrestricted access to your database.

Frequently asked questions