Let me start with a few numbers that left a deep impression on me.
In the latest version of Odoo 20, the R&D team has directly integrated something into the core product: native MCP server. Users just need to open the Enterprise edition, and the AI assistant can directly communicate with Odoo through the /mcp endpoint, without needing to set up an additional Python gateway or configure third-party middleware. Odoo itself is the MCP server.
On the same day, Odoo's paid user count surpassed 2.1 million—precisely 2,113,916. Of these, 67% were supported by the partner network. This milestone arrived ahead of schedule.
Looking at these two things together, they point in the same direction: open-source ERP is transforming AI from an "add-on feature" into a "native capability." And this time, it's not just on a PowerPoint slide — it has already been released.
From "Connecting an API" to "I Am the Server"
Let me translate what MCP is doing here.
The traditional way of connecting AI is like this: for your AI assistant to read Odoo data, you first need to build a middle layer — writing code, deploying services, handling authentication, and managing errors. What is this middle layer called? Some call it an "AI gateway," others call it an "Agent adapter." Whatever it's called, it's extra work.
Odoo 20's approach is: directly integrate the MCP endpoint into the product. AI connects via the MCP protocol, and Odoo handles permissions, data formats, and session management on its own. Developers do not need to write intermediate code or maintain a separate AI gateway service.
The user community's reaction was straightforward: "We've been waiting for this feature," "It will make competitors tremble again."
But I know what you care about is not how the community sees it—what you care about is what this means for your business.
It means your AI assistant can directly operate orders, inventory, and financial data in Odoo while inheriting Odoo's original permission system. It means you don't need to extract ERP data and put it on another AI platform to use it. It means the implementation cycle is shortened from "weeks" to "just a configuration."
Of course, currently this native MCP server mainly provides read permissions. Write operations and security policies are still being improved. But the direction is already very clear.
2 million paid users, 67% rely on partners
Another piece of data that made me look twice was that 67%.
2,113,916 Number of Odoo paid users (July 2026)
Odoo has 2,113,916 paying users, of which 1,422,084 (67%) are supported and serviced by the partner network. This proportion illustrates several things:
First, Odoo's ecosystem is much larger than many people imagine. Partners are not just selling licenses; they are doing real implementation and customization. Second, the "last mile" value that these partners provide to users is something pure SaaS vendors cannot offer. Third, when Odoo incorporates AI infrastructure like MCP into its core products, partners can directly use it in project delivery without having to learn an entirely new AI platform.
In other words: MCP entering Odoo core benefits not just users, but the entire ecosystem.
MCP: Becoming the "USB Interface" of ERP
This is not just Odoo's issue.
The MCP Enterprise Adoption Report, released in early July, shows that 78% of enterprise AI teams are already using the MCP protocol in their production environments. 28% of Fortune 500 companies are running their own MCP servers. Monthly SDK downloads reach 97 million. There are already over 9,400 public MCP servers on the market.
| MCP adoption metrics | Data | Time |
|---|---|---|
| Enterprises use MCP in production with AI teams | 78% | July 2026 |
| Fortune 500 companies run MCP servers | 28% | July 2026 |
| Monthly SDK Downloads | 97 million | July 2026 |
| Public MCP Server | 9400+ | July 2026 |
| AI vendors with native MCP support | Anthropic, OpenAI, Google, Microsoft | July 2026 |
And it's not just Odoo doing this. Workday DevCon has just released Agent-Ready Tools, also using the MCP protocol as the connection standard. In other words, mainstream ERP vendors—whether it's Odoo from the open-source camp or Workday from the enterprise SaaS camp—are all using the same protocol to enable AI Agents to communicate with ERP systems.
This is no coincidence. MCP is becoming the "USB interface" of the ERP world: a unified standard that allows any AI tool to plug in and work immediately.
To what extent has Agentification progressed
Having said that, I would like to share my assessment.
From a functional perspective, Odoo 20's MCP server is still in its early stages—primarily read operations, with security policies still evolving. But from a strategic perspective, this step is bigger than many realize.
The reason is simple: When MCP becomes a native capability of ERP, AI Agent is no longer "accessing ERP" but "living inside it."
This is an architectural difference. In the past, enterprises implementing AI-ERP integration would build an "AI shell" outside the ERP—an API layer, data pipelines, and independent security policies. Now Odoo has directly opened the door, allowing AI Agents to reside within the ERP, using the ERP's own data, permissions, and processes.
Workday's approach is another path—using Agent Passport for independent governance certification, but the underlying connection is also MCP. Different paths lead to the same destination.
Odoo becomes the MCP server itself, without any added layers.
—— Odoo R&D Team
Of course, this does not mean that Odoo users will be able to achieve fully automated ERP tomorrow. Native MCP solves the connectivity problem, but it does not solve the problems of governance, data quality, and business process redesign. These remain tough challenges in project delivery.
But the significance of what Odoo does lies in this: It turns the answer to the question "how to connect" into zero configuration. The remaining issues—how to govern, how to transform processes—are the old territory of ERP implementation, which partners and consultants are already good at.
A few things you can do right away
At the end, I have a few suggestions for friends who are using or considering Odoo:
First, if you are an Odoo Enterprise user, you can start using native MCP right now. Open your instance and see how the /mcp endpoint is configured. Start with small scenarios—such as having an AI assistant check order status or summarize inventory reports. Don't aim for full process automation right from the start.
Second, include MCP capabilities in your partner evaluation checklist. If your Odoo partner doesn't yet understand how to use MCP, now is the time for them to catch up. This capability will become a standard delivery configuration in the next 6-12 months.
Third, pay attention to the progress of Odoo's write permission MCP. The read operation MCP is already in place, and the timeline for implementing write operations is worth keeping a close eye on. Once AI agents can directly create orders, update inventory, and approve workflows in Odoo, many of your current implementation plans may need to be reconsidered.
Fourth, do not overlook ecological data. The combination of 2 million paying users and 67% partner share means that Odoo's ecosystem has grown too large to ignore. If you are selecting an ERP, this ecosystem scale itself is a consideration — more partners mean more choices, more mature services, and more reasonable prices.
