Model Context Protocol for Odoo.

What is Model Context Protocol?

Model Context Protocol is an open protocol that lets AI applications connect to external data, tools and workflows in a consistent way.

Without a standard such as MCP, every combination of AI model and business system becomes a separate integration. The interface from model A to system X then differs from model B to the same system, while switching to another AI platform requires new development work. MCP aims to reduce this integration problem by standardising the connection between AI applications and external systems.

MCP does not replace the underlying system. Odoo remains responsible for data, users, access rights and business logic. The MCP server forms a controlled translation layer between the AI client and the functions made available to that client.

Why does MCP exist?

A language model has no live access to an Odoo database by itself. It does not automatically know the current stock, outstanding invoices or a user's permissions. To work with up-to-date business information, the AI application needs a controlled connection to external systems.

Traditionally, such a connection is developed per application. A team defines endpoints, tool descriptions, authentication, error handling and mappings for every model or agent platform. That works, but quickly leads to multiple variants of the same integration.

MCP introduces a common layer. As a result, an MCP-compatible client can work with multiple servers and a single server can make its capabilities available to multiple compatible clients. The exact support and functionality still depend on the chosen client and implementation.

What MCP does and does not solve

How does the MCP architecture work?

MCP uses a host, client and server model. The user works in an AI application; the client maintains the protocol connection; the server exposes controlled functions and context.

The host is the application in which the user works with the AI model. The host can contain one or more MCP clients. Each client maintains a connection with an MCP server. The server describes which capabilities are available and processes requests that fall within the permitted policy.

Tools, resources and prompts

Not every implementation needs to use all primitives. For an enterprise Odoo integration, the emphasis is usually on controlled tools and structured results. Which functions are published should be an explicit design decision.

How does MCP work with Odoo?

In the DAADit architecture, the MCP server sits outside the Odoo user interface. The server receives standardised MCP requests, checks the identity and the configured policy, and translates a permitted action into an Odoo operation.

Example of a tool definition

// One of the tools your AI client receives after tools/list.
{
  "name": "odoo_search_read",
  "description": "Search and read records in a single call — the most-used tool for questions about your data.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "model": { "type": "string" },
      "domain": { "type": "array" },
      "fields": { "type": "array" },
      "limit":  { "type": "integer" }
    },
    "required": ["model"]
  }
}

The full toolset of the hosted server consists of odoo_list_instances, odoo_list_models, odoo_fields_get, odoo_search, odoo_search_read, odoo_read, odoo_create, odoo_write and odoo_execute_kw. Which of these a client actually sees is determined per environment and per user with capability gating; by default only the read-only tools are enabled.

The server does not determine on its own what the user means. The AI client selects a tool based on the description and the interaction. That is why tool names, schemas, error messages and authorisation must be carefully designed and tested.

MCP versus a traditional REST API

MCP and REST are not direct alternatives. An MCP server often uses existing APIs or system methods to communicate with the source system. The difference lies mainly in the interface towards the AI application.

An important architectural principle is that MCP must not bypass existing authorisation. The MCP layer should respect the permissions in Odoo and can additionally apply stricter scopes.

Security, privacy and governance

An MCP implementation is not secure simply because the protocol is used. Security comes from identity, least privilege, explicit boundaries, controlled tools, logging and operational management.

Model gating and capability gating

Model gating determines which AI clients or models may access a tenant or functional area. Capability gating restricts which categories of actions are available, for example reading, searching, creating, modifying or executing specific methods.

DAADit applies these concepts as product policy on top of the existing Odoo permissions. A user can never be allowed more than the linked Odoo identity, while the MCP layer can restrict access further.

Logging and audit trail

For management and governance, it must at minimum be visible which user or service identity started a call, which tool was chosen, which tenant and environment were involved, whether the action was permitted and whether it succeeded. Sensitive prompt content and results require a separate retention policy.

Logging is not a substitute for preventive security. It forms the basis for incident investigation, usage analysis, troubleshooting and improving tool definitions.

Implementation of DAADit Hosted MCP

The implementation does not start with an AI model, but with the desired business processes and the risk class of the associated actions. An information request about stock requires a different level of control than creating or confirming financial entries.

Recommended implementation phases

Dashboard and support

The customer gets a compact dashboard for connections, users, models, capabilities and activity. The complex server configuration remains under DAADit's management. When a user reports a problem, the helpdesk can investigate the relevant environment, logging and configuration. A support question can be forwarded to DAADit from the same AI assistant.

Examples of Odoo MCP applications

The examples below illustrate different levels of reading, analysing and executing. The final toolset is designed per customer and persona.

More extensive business stories can be included in the section Put Odoo to work. This product page deliberately stays focused on the technical workings and administration.

Related documentation and articles