In early July, news spread in the Odoo community that enterprise users can now directly use the /mcp endpoint to allow AI assistants to read and write Odoo data. No need to set up external gateways or install third-party middleware—native support.
On the same day, a partner posted data on LinkedIn: Odoo's paid users officially exceeded 2.11 million, with 67% served by the partner network.
Looking at these two things together, they are actually talking about the same thing: open-source ERP is answering an industry-level question in its own way—when AI wants to "integrate" into the core of enterprise management, what kind of architecture can handle it?
At the same time, an analysis article from Enterprise Network D1Net provides context: in 2026, ERP is undergoing structural changes, with AI beginning to take over high-frequency tasks such as invoicing, reconciliation, and onboarding, while integrated systems are starting to "loosen."
Native MCP – Odoo’s AI Interface Strategy Takes a Different Path from Big Tech
MCP (Model Context Protocol) is an open protocol launched by Anthropic last year, designed to establish a standardized communication method between AI models and external tools. This thing exploded rapidly in the first half of this year—Anthropic, OpenAI, Google, and Microsoft all natively support it, with the MCP SDK approaching 100 million monthly downloads.
In the ERP field, Odoo is the first to integrate MCP into its core product. Not through plugins, not through third-party adapters, but by directly embedding an MCP server in Odoo 20 Enterprise Edition, exposing a /mcp endpoint. Once the AI assistant enters, which tables to read, which fields to write, which methods to call—all go through this standardized channel.
I took a careful look at the technical details, and there are a few points worth mentioning:
First, zero external dependencies. No need to install additional Python gateways or Node.js adapters. Odoo itself is the MCP server. This saves a lot of trouble for implementers and operators—one less component means one less point of failure.
Second, the security and permission layer is handled within Odoo. MCP requests do not bypass Odoo's original ACL and Record Rules. What AI can do follows the same permission system as what human users can do.
Third, enterprise edition exclusive, but there are also community solutions. Native MCP is currently an enterprise edition feature. However, community edition users are not completely without options — in the first half of this year, multiple third-party MCP-Odoo adapters have emerged in the community, with one even appearing on the Tencent Cloud Developer Platform, written in Java with 80,000 lines of code.
2,113,916
Number of Odoo paid users (July 2026, target achieved ahead of schedule) 67% of users served by the partner network
Integrated ERP is loosening—Is AI "demolishing walls" or "reinforcing" them?
In early July, enterprise network D1Net published an analysis with a very straightforward title: "ERP in 2026: AI Takes Over Processes, Integrated Systems Begin to 'Loosen'." I read it carefully, and the core argument is not that ERP is dying, but that its form is splitting.
The article cited interviews with several CIOs, and their statements were surprisingly consistent:
"The ERP market is shifting from a pure transaction system to an intelligent, data-driven platform. AI and predictive analytics have been integrated into core processes, with tighter integration with data lakes, and flexible modular architectures enable enterprises to quickly adapt to changes without the need for complete overhauls as in the past." — Steve Bronson, CIO of Southern Glazer's
This sentence contains a hidden thread: modularity. For decades, ERP vendors have been selling "all-in-one packages" — finance, supply chain, HR, CRM, and BI all bundled into one system. The advantage is high integration, but the downside is that modifying one module means affecting the entire system. Now, AI offers enterprises another option: use AI agents to extract core processes and build an "intelligence layer" outside the core ERP, handing over high-frequency, repetitive, and rule-based tasks to the agents, while humans step back into roles of review and decision-making.
Gartner also provided a timeline: by the end of 2026, 40% of enterprise applications will have built-in AI agents. All major ERP vendors are moving in this direction.
The question is—are you running in the right direction?
Comparison of Two Paths
| Dimension | Traditional integrated ERP + AI add-on | Modular ERP + AI Native Integration |
|---|---|---|
| AI access method | Plug-in or standalone AI platform | Native MCP/API, system as interface |
| Permission Management | Multiple permission systems, complex operations and maintenance | Unified ACL, AI and humans share the same permissions |
| Knowledge Ownership | Easy flow to vendor shared model | Data belongs to the customer, code is auditable |
| Customization flexibility | Limited by vendor module boundaries | Open source, deeply customizable |
| License cost | By seat, AI access cost is high | By application/user, open source lock-free |
Of course, integrated ERP won't disappear overnight. AppDirect's CIO put it bluntly: "Integrated systems will continue to exist, but vendors need to rethink their pricing. More and more companies no longer need an all-in-one package where 90% of the features are never used—if you're going to sell it that way, don't price it that way."
Where does Odoo stand in this diagram
Odoo's strategy is quite different from traditional ERP vendors. Instead of first building a large, comprehensive system and then adding AI on top, it first creates a modular, detachable and combinable system, and then uses open protocols like MCP to allow AI to directly "read" and "write" — not through interface simulation, but through structured data channels.
The logic behind this is essentially what Gartner recently referred to as "agent arbitrage"—when AI agents become the primary users of enterprise software, what you need is not a visually appealing interface, but a well-structured data model and standard API channels. Under this premise, Odoo's modular architecture, unified data model, and open-source license indeed offer structural advantages.
But to be honest: Odoo's AI capabilities are currently mainly focused on the layer of "enabling AI to read and write data." The true "intelligence layer"—such as autonomous decision-making, anomaly alerts, and process orchestration—still needs to be built by the implementer, or wait for further maturity from the community and official sources. Odoo does not have 200 pre-built intelligent agents like SAP, nor has it directly launched an "Lingji" AI operating system like Kingdee. Odoo's approach is: give you the key, and you open the door yourself.
This key is the native MCP.
Several Practical Suggestions for Enterprise CIOs
At the end of the article, I want to say a few practical things.
If you are selecting or upgrading an ERP system and have started considering AI integration, these judgments can serve as a reference:
First, don't just look at the UI; look at the API first. No matter which ERP you choose, AI integration will become a necessity in the next two to three years. If your ERP lacks a standard, comprehensive API channel, or if the contract terms do not allow autonomous calls by intelligent agents—you will be in a very passive position later on.
Second, modularity and openness are more valuable than "integration". The all-in-one ERP model may become increasingly difficult to sustain in the AI era — because you don't need to replace everything, only to overlay AI capabilities on key processes. A system that can be disassembled and assembled, with an open data model, will be far more flexible than one where "any change requires overhauling the whole thing."
Third, open-source licenses are a hidden advantage in the AI era. No per-user fees, no restrictions on API calls, and customer ownership of data—these "unremarkable" terms become tangible cost and compliance advantages when AI begins to integrate with ERP at scale.
