Connect CrewAI to Odoo
via MCP

Why connect CrewAI to Odoo?

CrewAI is a framework for orchestrating multi-agent systems — teams of autonomous agents, each with a role and a goal, collaborating to complete a task. The moment those agents need to reason over your real business — open sales orders, unpaid invoices, stock levels, CRM pipeline — they need live access to Odoo.

Wiring each agent to Odoo by hand with bespoke XML-RPC calls is brittle and hard to govern. A cleaner approach is MCP (Model Context Protocol): expose Odoo once as an MCP endpoint, and any MCP-aware client — including CrewAI — can consume it as a set of structured, permission-scoped tools.

This guide shows how to give a CrewAI crew live Odoo access through MCP, with a realistic code sketch, two hosting paths, and the security model that keeps agents inside their lane.

The approach: CrewAI consumes Odoo MCP tools

CrewAI supports MCP tools through the MCPServerAdapter shipped in the crewai-tools package. You point the adapter at an MCP server, it discovers the tools that server advertises, and CrewAI treats them as first-class agent tools — the agent decides when to call each one.

On the Odoo side, the MCP endpoint is exposed by the KSROlabs MCP Server. It advertises tools such as search_records, get_record, and describe_model, each scoped to the models your API key allows. Because MCP is an open standard, the same Odoo endpoint you build for CrewAI also works with Claude, Cursor, and any other MCP client.

Step-by-step: give a CrewAI crew Odoo access

Follow these 5 steps to connect a CrewAI crew to live Odoo data via MCP.

Put together, a minimal CrewAI + Odoo MCP script looks like this:

from crewai import Agent, Task, Crew
from crewai_tools import MCPServerAdapter

# 1. Point the adapter at your Odoo MCP endpoint
odoo_mcp = {
    "url": "https://your-odoo.com/mcp/v1",
    "headers": {"Authorization": "Bearer <YOUR_SCOPED_TOKEN>"},
}

# 2. Load the Odoo MCP tools and hand them to an agent
with MCPServerAdapter(odoo_mcp) as odoo_tools:
    analyst = Agent(
        role="Revenue Analyst",
        goal="Answer questions using live Odoo ERP data",
        backstory="An operations analyst who reasons over Odoo records.",
        tools=odoo_tools,          # search_records, get_record, describe_model, ...
        verbose=True,
    )

    task = Task(
        description="List this week's unpaid invoices and flag the top 5 by value.",
        expected_output="A ranked list with customer, amount, and due date.",
        agent=analyst,
    )

    crew = Crew(agents=[analyst], tasks=[task])
    result = crew.kickoff()
    print(result)

Illustrative sketch — adapt the URL, token, models, and task to your own Odoo instance and crew design.

Two ways to expose your Odoo MCP endpoint

The right hosting path depends on how your Odoo is deployed. CrewAI connects the same way to both.

For self-hosted setups, see the Odoo MCP Server module. For SaaS instances, the Odoo Online MCP Gateway is the path for connecting CrewAI to an *.odoo.com deployment.

Use cases and security

Once a crew has grounded Odoo access, multi-agent automation over your ERP becomes practical:

Because permissions live in Odoo — not in your agent code — you can hand a crew broad reasoning ability while keeping a tight boundary on what it can actually read or change. Start read-only, grant write access to individual models only when a workflow needs it, and issue a separate key per crew or environment so you can revoke one without disturbing the rest.

Frequently asked questions