Odoo 20 AI: Agents, MCP and Who Stays in Control
Odoo 20 lets AI agents act on automation rules and lets ChatGPT or Claude connect through MCP. What that means for productivity, control and risk.
This is part 6 of our Odoo 20 series. AI is the area of Odoo that grew fastest in this release: the number of AI-related modules went from 15 in Odoo 19 to 51 in Odoo 20. Rather than list them, this part answers the questions management actually asks: what can it do for us, and how do we stay in control?
From AI assistant to AI agent
Odoo 19 introduced AI that helps a person who is using the screen: drafting text, reading documents, answering questions. Odoo 20 takes the next step. AI agents can now be triggered by automation rules.
An automation rule is the familiar "when X happens, do Y" setting in Odoo. Y can now be "let an AI agent handle it". Some illustrative uses:
- When a supplier bill arrives, an agent reads it and prepares the draft entry for review.
- When a customer sends a helpdesk ticket, an agent classifies it and drafts a first reply.
- When a sales opportunity goes quiet, an agent prepares a follow-up for the salesperson to approve.
Odoo 20 also adds AI building blocks across many apps, including accounting reports, purchasing, sales, inventory, projects, helpdesk, marketing automation, social media and website live chat. Agents can be equipped with reusable skills and tools, and their conversations are tracked as sessions.
The management question is not "can AI do this?" but "where do we want a human to approve before anything is final?" Start with tasks where the agent prepares and a person confirms.
Odoo 20 speaks MCP
The second development is less visible but strategically important. Odoo 20 Enterprise includes an MCP server.
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to business systems. In practice it means someone could ask an assistant such as ChatGPT or Claude, "Which customers are more than 60 days overdue?", and the assistant fetches the answer from Odoo instead of guessing.
What matters for control is how Odoo does this:
- Nothing is exposed by default. An administrator chooses which actions the AI may use. Each one is a specific, named action, not open access to the whole database.
- Each action is marked read-only or not. The connecting AI assistant is told which actions only look and which ones change data.
- Normal permissions still apply. The AI acts with the access rights of the user who connected it. It cannot see what that user cannot see.
- Sign-in is controlled. Users connect their AI assistant through an approval step, with a pre-approved list of known assistants (including ChatGPT, Claude and Grok). The access key used for AI is separate from keys used by other integrations.
What MCP is, and isn't
MCP in Odoo 20 is built for "ask the ERP a question" and for carefully chosen actions. It is not a general-purpose connection for your other applications. A mobile app, a webstore or a delivery platform that needs to read and write data across Odoo still needs a proper integration.
The two fit together. MCP is a curated door for AI assistants. Your other systems need a controlled door of their own. At Keyklik we built ODXProxy for that second door: a gateway that lets ordinary applications work with Odoo through a short, fixed list of permitted operations, with their own separate keys. Using both, AI gets the handful of actions you approved, and your apps get exactly the data access they need, no more.
A sensible way to start
- Pick one repetitive task where the AI prepares and a person approves. Measure time saved after a month.
- Expose read-only questions first through MCP (overdue receivables, stock levels, open orders) before allowing any action that changes data.
- Decide who owns the AI setup: who approves new actions, who reviews AI sessions, and who switches things off if needed.
- Write it down. A one-page policy on what AI may and may not do in your ERP is worth more than any feature.
- Check the data question. Connecting an external AI assistant means business data leaves Odoo to be processed elsewhere. Make sure that fits your contracts and data-protection obligations.
Want to explore AI in Odoo 20 safely? See our services or contact the Keyklik team.
The Odoo 20 series
- What's New for Your Business
- Point of Sale for Restaurants and Retail
- Inventory and Manufacturing
- Accounting and Purchasing
- HR and Attendance
- AI Agents and MCP (this article)
- Planning the Upgrade
Note (DYOR — Do Your Own Research): This article is based on our own comparison of the Odoo 19 and Odoo 20 source code, not on Odoo's official release notes. AI features, editions and the list of supported assistants can change. AI output can be wrong and should be reviewed before it affects financial, tax or customer records. Sending business data to external AI services raises data-protection questions (including under the PDP Law). Verify against the official Odoo documentation and your AI provider's terms, and consult your legal and IT advisors before connecting AI to your ERP.