Odoo 20 AI Agents and MCP: What They Can Do and How to Control Them

Odoo 19 introduced an AI app. Odoo 20 pushes further, according to summaries of its release notes. The agents can now do real work inside your ERP, which makes it worth understanding what they can do and how to keep them under control.

Source note: the details below come from a partner's summary of the official Odoo 20 release notes. Confirm them in Odoo's own documentation before you plan around them.

What the agents are reported to do

• Create and update records, including from uploaded files such as PDFs. A supplier bill can become a draft vendor bill without retyping.
• Run from automated or scheduled actions, so an agent can work in the background instead of waiting for someone to open a chat.
• Show live progress while they work.
• Ask for confirmation before exceeding tool-call limits.
• Connect to outside tools through MCP, the Model Context Protocol, an open standard for linking AI systems to tools and data.
• Choose a model for each task automatically, once you have selected an AI provider.

AI features reportedly require purchased IAP (in-app purchase) credits, so usage has a cost.

Practical use cases

1. Vendor bill capture: read a PDF, create a draft bill and match it to a purchase order.
2. Lead intake: turn enquiry emails into CRM leads with the right fields.
3. Scheduled checks: every morning, list overdue invoices or stock below reorder level and prepare a summary.
4. Knowledge help: answer staff questions about how to perform a task in Odoo.

Guardrails to set before switching on

• Start read-only, then allow drafts, then allow posting only with approval.
• Use a dedicated user for each agent, with only the access it needs.
• Require human approval for anything that sends email, posts accounting entries, changes prices or pays a bill.
• Treat uploaded files as untrusted. A PDF can contain hidden instructions, so never let the agent change payment status or bank details on its own.
• Be careful with MCP. Each connected tool widens what an agent can reach, so connect only tools you trust and review what each one can do.
• Set a budget and watch credit usage.
• Keep logs so you can see what the agent did and undo it.

Test on real data first

Run the agent against 30 to 50 past documents on a staging database. Count the errors, note which types of document it struggles with and decide what still needs a person.

How KalzTech helps

We configure Odoo AI features and build custom agents with approval steps, permissions and logging. If you want to try one on a real workflow, get in touch.