Odoo MCP is the use of the Model Context Protocol (MCP) to give AI assistants controlled access to an Odoo database. An Odoo MCP server publishes tools such as "search partners", "list overdue invoices" or "create a draft quotation". Each tool calls Odoo's external API as a specific Odoo user, so Odoo's access rights and record rules still apply. The assistant decides which tool to call; the server decides what the tool is allowed to do.
What is the Model Context Protocol?
The Model Context Protocol is an open specification for connecting AI applications to external systems. The current version is 2026-07-28; earlier versions were 2025-11-25, 2025-06-18 and 2025-03-26 (versioning page).
The spec defines three roles:
- Host: the AI application the user works in, such as Claude Desktop or ChatGPT.
- Client: the connector inside the host that talks to one server.
- Server: the program that exposes capabilities, here the Odoo MCP server.
Servers offer three kinds of primitives: tools (functions the model can call), resources (data the application can read) and prompts (templates for the user). Messages use JSON-RPC 2.0. Transports are stdio, for a server running on the same computer, and Streamable HTTP, for a remote server. Remote servers can use OAuth-based authorization defined in the spec.
In December 2025 Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation.
How does an Odoo MCP server map tools to Odoo?
Odoo exposes every model through its external API: search_read, read, create, write, read_group and the model's public methods. An MCP server chooses which of these become tools and how they are described.
| MCP tool (example) | Odoo model | Odoo method | Type |
|---|---|---|---|
search_partners |
res.partner |
search_read |
Read |
overdue_invoices |
account.move |
search_read with a domain on due date and payment state |
Read |
stock_on_hand |
stock.quant |
read_group by product and location |
Read |
partner_ledger |
account.move.line |
search_read filtered by partner and account type |
Read |
create_draft_quotation |
sale.order |
create with state draft |
Write |
add_lead_note |
crm.lead |
message_post |
Write |
There are two design styles:
Generic tools
A few tools such as search_read, create and execute_kw that work on any model. Flexible for developers, but the model must know Odoo's data model, and any method the user can call becomes reachable.
Task-shaped tools
Named tools such as "overdue invoices" or "draft quotation" with fixed domains and validated inputs. Fewer surprises, easier to review and to secure, but each new task needs a new tool.
Many servers mix both: task-shaped tools for common questions and a generic read tool for everything else, with writes limited to an allow-list.
Which Odoo API does it call?
Servers for Odoo 16 to 18 use XML-RPC or JSON-RPC with an API key in place of the password. Odoo 19 adds the JSON-2 API at /json/2/<model>/<method> with a bearer API key. Odoo states that XML-RPC and JSON-RPC are scheduled for removal in Odoo 22 (fall 2028) and Odoo Online 21.1 (winter 2027).
On Odoo Online and Odoo.sh subscriptions, external API access is available only on the Custom plan, according to the same documentation. This applies to every MCP server that connects from outside.
Does Odoo have MCP natively?
As of 20 October 2026:
| Odoo version | Native MCP server |
|---|---|
| Odoo 16, 17, 18 | No |
| Odoo 19.0 | No |
| Odoo Online saas-19.4 and later | Yes, as part of Odoo AI (Enterprise) |
| Odoo 20 (released September 2026) | Yes, as part of Odoo AI (Enterprise) |
| Odoo Community, any version | No |
The Odoo 20 MCP documentation describes an endpoint at <database URL>/mcp, sign-in through OAuth or an API key with the MCP scope, and tools defined as server actions. The Odoo 20 release notes list it as "Connect Odoo to anything: connect to your database via MCP". The same release notes state that all AI features now need credits bought through IAP; whether that applies to MCP use specifically is not stated on the MCP page.
So the honest answer to "does Odoo have MCP" is: yes, if you run Odoo 20 or recent Odoo Online with the AI app; no, if you run Odoo 16 to 19 on-premise or Community, which is where most existing installations are. For those, open-source and hosted servers fill the gap. See the comparison.
MCP versus Odoo's built-in AI
Odoo's AI app, since Odoo 19, runs agents inside Odoo with OpenAI or Gemini models configured in Odoo. MCP works the other way round: the AI assistant your team already uses calls into Odoo. Both can coexist. The trade-offs are covered on why Odoo MCP.
What makes an Odoo MCP server safe?
- it acts as a dedicated, least-privilege Odoo user, ideally one per person;
- writes are off by default, allowed only for named actions, and need confirmation;
- changes can be undone and every call is logged;
- delete is not exposed.
Details are on the security page. To try it, follow the setup guide.
Connect AI to your Odoo
Describe your situation in a few lines. We reply by e-mail.
Frequently asked questions
Does Odoo have MCP?
Odoo 20, released in September 2026, and Odoo Online from saas-19.4 include an MCP server as part of Odoo AI, which is an Enterprise feature. Odoo 19.0 and earlier do not have one, and Odoo Community has none. For those versions, open-source and hosted Odoo MCP servers connect through the external API.
What is the difference between an MCP server and Odoo's external API?
The external API is a generic way to call Odoo model methods. An MCP server sits on top of it and describes a set of tools with names, descriptions and input schemas, so an AI assistant can discover and call them without custom integration code for each assistant.
Which AI assistants support MCP?
MCP is supported by Claude on web, desktop and Claude Code, ChatGPT through custom MCP servers, Cursor, VS Code with GitHub Copilot, Microsoft Copilot Studio, Gemini CLI and Mistral, among others. Support for remote servers, sign-in and write actions differs by client and plan.
Is MCP an Anthropic-only standard?
No. Anthropic introduced MCP and in December 2025 donated it to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. The specification and SDKs are open source.