Odoo 20 and MCP: Connecting AI to Your Odoo ERP

There has been a lot of discussion about AI in ERP over the last two years, but one of the Odoo 20 developments that caught our attention at AP Systems is actually quite technical.

It's called MCP, or Model Context Protocol.

MCP gives compatible external AI applications a structured way of connecting to Odoo. Instead of the AI sitting outside the ERP with no knowledge of what is happening in the business, it can potentially use tools made available by Odoo to work with information in the database.

For us, that's far more interesting than simply adding another chatbot.

The real opportunity is giving people an easier way to work with the information already sitting inside Odoo.

What Does MCP Actually Do?

A normal AI model doesn't automatically know what's happening in your Odoo database.

It doesn't know which quotations are outstanding, which Helpdesk tickets haven't been resolved or which purchase orders are overdue.

You could give it some of that information manually, or build a dedicated integration to provide it, but that's not particularly practical if you want AI to become part of everyday business processes.

MCP takes a different approach.

With Odoo acting as the MCP server, a compatible AI application can connect to Odoo and use specific tools that have been made available to it.

That last part is important.

The idea isn't to simply open the entire Odoo database to an AI platform. You decide what functionality is exposed, and the connection operates within the permissions of the Odoo user involved.

From an integration point of view, this is what caught our attention.

We already spend a lot of time connecting Odoo to systems that each have their own APIs, authentication methods and business rules. MCP doesn't replace those integrations, but it gives us a different way of approaching AI.

What Could You Actually Use It For?

Imagine a sales manager asking:

"Show me our quotations over R100,000 that haven't been confirmed."

Then:

"Which of those customers haven't been contacted in the last seven days?"

That's a fairly normal business conversation.

Today, the manager might open Odoo, find the correct menu, apply filters, group the results and then work through the records.

An experienced Odoo user will probably do that quite quickly.

Someone who only uses Odoo occasionally may ask another employee to prepare the information for them.

With an MCP-connected AI application, the manager could potentially ask the question in normal language and have the AI use the appropriate Odoo tools to retrieve the information.

The underlying data is still in Odoo. The AI simply gives the user another way of working with it.

That's where we see the opportunity.

Helpdesk Is a Good Example

Helpdesk is one area where we think this could become useful quite quickly.

A service manager could ask:

"Which high-priority tickets have been open for more than two days?"

Once those have been identified:

"Which ones are waiting for us rather than the customer?"

And finally:

"Give me a short summary of each ticket."

The manager isn't particularly interested in which Odoo model or field contains the information. They want to know where the problems are and which customers need attention.

AI is well suited to that type of interaction.

The same idea could apply to CRM, purchasing, inventory and other areas where managers regularly need to investigate exceptions rather than process every individual transaction themselves.

Manufacturing Has Some Interesting Possibilities

Manufacturing is another area we'll be looking at closely.

A production manager might want to ask:

"Which manufacturing orders are currently delayed because we don't have the components?"

Or:

"Which components have caused the most production shortages this month?"

There are already ways to investigate this information in Odoo, and MCP doesn't suddenly make conventional reporting unnecessary.

What it could do is make information more accessible.

A production manager shouldn't necessarily need to understand every Odoo filter or reporting option to ask a sensible question about the factory.

If AI can help get from that question to the relevant Odoo information more quickly, there is a practical benefit.

Reading Information Is One Thing. Changing Odoo Is Another.

This is where we'd be careful.

Odoo's MCP functionality isn't necessarily limited to retrieving information. Depending on the tools exposed and the permissions of the user, an external AI agent can potentially perform actions that modify data.

Consider these two requests:

"Show me the sales opportunities that haven't been followed up."

and

"Create follow-up activities against all of those opportunities."

The first request reads information.

The second changes Odoo.

We wouldn't treat them in the same way.

There are plenty of actions where AI assistance could make sense. Creating an activity, preparing information or helping classify a Helpdesk ticket may be relatively low risk.

We'd take a very different approach to allowing an AI agent to adjust stock, change customer credit limits, confirm significant purchase orders, post accounting transactions or approve payments.

Just because the technology can perform an action doesn't mean it should be given permission to do so.

Odoo already has access rights and approval processes for a reason. AI needs to work inside those controls, not around them.

Security and POPIA Need to Be Part of the Design

This becomes particularly important when connecting an external AI service.

The business needs to understand what information the AI platform can access, what information is being processed outside Odoo and whether customer, employee, supplier or other personal information is involved.

For a South African business, POPIA therefore needs to be considered when designing the integration.

We wouldn't describe an AI integration as "POPIA compliant" simply because an API key and Odoo permissions are being used.

There is more to it than that.

You need to look at what data is being processed, why it is required, who has access to it, where it goes and what contractual and security arrangements apply.

Our preference would also be to start with the minimum access required.

If an AI application only needs CRM information to solve a particular problem, there is no reason to give it access to Accounting, Payroll or other areas of the database.

MCP Doesn't Mean Everything Should Become AI

This is something we think will become increasingly important as AI functionality expands.

There is no point replacing a simple Odoo process with a complicated AI process just because the technology is available.

A warehouse operator scanning a delivery doesn't need to have a conversation with an AI agent.

An accountant capturing a transaction may be better served by the normal Odoo interface.

Where AI starts making more sense is when someone has to search, investigate, summarise or work across a lot of information.

That's why the sales manager, Helpdesk manager and production manager examples are interesting.

They're asking questions.

MCP potentially gives the AI a controlled way of getting the information required to answer them.

How We Would Approach an Odoo MCP Project

We wouldn't start by connecting an AI platform to the whole database and then deciding what to do with it.

We'd do the opposite.

Find one business problem first.

For example:

Our sales manager spends too much time identifying quotations that need attention.

Then determine what information the AI actually needs from Odoo to solve that problem.

Give it access to that information.

Test the results.

Check the permissions.

Make sure users understand what the AI is doing.

Then measure whether it actually saves any time.

If it works, move on to the next use case.

This is generally how we prefer to approach automation anyway. Solve something small and useful before trying to automate half the company.

Why We're Watching MCP

At AP Systems, we've spent years integrating Odoo with external platforms — eCommerce systems, marketplaces, couriers, payment providers, EDI networks and other business applications.

MCP doesn't replace any of that.

What it does provide is a potentially more standard way for AI applications to work with functionality made available by Odoo.

That could become quite important.

A business may use Odoo as its ERP but prefer a particular external AI platform for certain tasks. Another business may use several AI tools for different purposes.

MCP potentially makes it easier for those worlds to meet.

We still think Odoo should remain the system controlling the business process.

Customers, products, stock, quotations, purchase orders, manufacturing orders and accounting transactions belong in the ERP.

AI doesn't need to replace that.

It needs to make it easier for people to work with it.

That's why MCP is one of the Odoo 20 developments we'll be testing closely at AP Systems.

Not because every customer suddenly needs an MCP project, but because it could solve some very practical problems around accessing information and automating repetitive work.

The first question shouldn't be:

"How do we connect AI to Odoo?"

The technology is increasingly giving us ways to do that.

The better question is:

"What should we allow it to do once it's connected?"

That's where the real implementation discussion starts.