Odoo AI agent for your team

What an Odoo AI agent does in a normal day

Most Odoo questions are small and frequent. How many invoices are unpaid? Which orders are waiting on a transfer? What was said on that quote last week? The Odoo Agent answers them in a chat, so nobody has to open five lists and build a filter.

It also does the work. After a change it reads the record back and reports any difference, so you see what really changed and not what it hoped to change.

• Search, read, and count records on any model the employee can access: sales orders, invoices, contacts, products, transfers, and leads
• Totals and rankings, such as the top five customers this quarter
• Create and update records, and press the same buttons your team does, such as confirm an order or post an invoice
• Guided wizards: send an email, register a payment, or reverse an entry
• Read the chatter on a record and summarise what happened
• Odoo PDF reports and quick insight PDFs, served through expiring download links

It acts as the employee who asked

Every request uses that employee’s own Odoo login or API key. Odoo’s access rights and record rules apply as usual. Where you are blocked in Odoo, the agent is blocked too. There is no shared admin user doing the work behind the chat.

Writes to access rules, groups, and similar security models are blocked at the platform level, and sensitive fields are protected. Requests for passwords and API keys are refused. Every tool call is logged with the workspace, the Odoo user, the tool, the model, and the record ids. Connector credentials are encrypted at rest.

Odoo MCP for Claude and Cursor

The same tools are available over MCP, the open standard that lets an AI client call tools. Connect Claude or Cursor with your workspace MCP key plus the employee’s Odoo login headers, and the editor or assistant works with Odoo as that person. A REST API exposes the same tools over HTTP for your own scripts.

This is a staff tool. The read-only commerce MCP, which lets an assistant search a store’s products and stock, is a different endpoint on the eCommerce Agent. The AI chatbot for Odoo guide explains the split between the shopper bot and this one.

How to set up the Odoo AI agent

Your workspace needs a connected Odoo 19 database, and each person acts with their own access. The Odoo Agent works with Odoo Online, Odoo.sh, and self-hosted installs.

1. 1

   Connect Odoo

   Authorise with the doo.mcp Connect add-on or an Odoo API key. The add-on opens an Odoo login in a popup and creates a scoped API key, so the dashboard never stores a password.

2. 2

   Each person signs in once

   Employees use their own Odoo login or API key. That is what makes the agent act as them. The workspace connector user is only a fallback when someone has not added their own key.

3. 3

   Ask in the dashboard chat

   Open the Odoo Agent in your workspace. Keep separate threads, pin the ones you return to, and click a record name in an answer to open it in Odoo.

4. 4

   Connect Cursor or Claude over MCP

   Add the MCP server with your workspace key and the employee’s Odoo login headers. The tools you see are the tools the dashboard chat uses.

5. 5

   Share the workspace link

   Staff can open the agent at your workspace address and sign in with their Odoo login. Anyone with the address reaches the sign-in page, but only a person with a valid login for your database can use the agent, so treat it like an internal tool.

A sample conversation, from a question to a posted invoice

Each grey tag is a real call made as the signed-in employee. If that employee could not post invoices in Odoo, the last step would fail, and the agent would say so.

1. EmployeeHow many unpaid customer invoices do we have?

2. count_recordsaccount.move, unpaid customer invoices

3. Odoo Agent47 unpaid customer invoices.

4. EmployeeTop 5 customers by amount, and send me a PDF.

5. aggregate_recordsGrouped by customer, summed
6. generate_insight_pdfunpaid_invoices, 42 rows

7. Odoo AgentHere are the top five. Your PDF is ready to download.

8. EmployeePost invoice INV/2026/00123.

9. execute_methodaction_post

10. Odoo AgentPosted. The invoice is now in the posted state.

What beta means here

Beta means the agent is live and usable, and these gaps are known. Read the list before you give it to a team that works with invoices and stock every day.

• No built-in confirmation step before deletes and bulk edits yet. Deletes are limited to a handful of records per request and follow the employee’s permissions.
• Posting messages and scheduling activities from the chat is not released.
• Odoo 19 only. Odoo 16, 17, and 18 are planned.
• It works from the dashboard, over MCP, and over REST. An assistant inside Odoo’s own interface is on the roadmap.
• Signing in with OAuth, instead of an Odoo login or API key, is on the roadmap.

What the Odoo Agent costs

There is no free plan. The Odoo Agent is paid from the first plan, and each agent is its own subscription. Paid checkout is opening soon, and the prices on the pricing page are the ones that will be charged.

• Starter, $49 a month: 25,000 AI responses, each person on their own Odoo access, and MCP for Cursor and Claude.
• Growth, $99 a month: 150,000 AI responses.
• Scale, $199 a month: 1,000,000 AI responses.
• Business, $349 a month: 1,000,000 AI responses.

Each reply the agent writes, or each action it takes such as a search or an Odoo update, counts as one AI response. The cap belongs to the Odoo Agent. If you want something free to try first, the eCommerce Agent and the website chatbot both start free.

How it differs from Odoo’s own AI

Odoo 19 ships its own AI app, with agents you configure inside Odoo and an Ask AI prompt for people already in the database. The Odoo Agent is a separate product. It works from the doo.mcp dashboard and from Claude and Cursor over MCP, and it connects to your database with each employee’s login or API key. You can read Odoo’s description of its own AI in its documentation.

Sources: Odoo 19 documentation: AI.