Odoo MCP Server: Connect AI to the Open-Source ERP

What is an MCP server for Odoo?

Odoo is probably the most popular open-source ERP among mid-sized companies in German-speaking countries: a modular toolkit that holds everything together in one database, from sales through accounting and inventory to manufacturing. In everyday use, though, this very breadth is also the brake: the number you need is three clicks deep in some module.

An MCP server for Odoo puts a translator in between: it accepts the structured tool calls of an AI agent and converts them into Odoo RPC calls. The server handles authentication, field selection, permission checks and formatting, and the Odoo records reach the model as cleanly readable text instead of raw XML. This lets you search, create and update Odoo in natural language.

Important context: we are not aware of an “official” MCP server published by Anthropic or by Odoo SA itself. There is a lively community ecosystem, including a widely used Python package on PyPI and a module in the Odoo App Store. Where such a building block fits, we use it; where not, we build the connection directly on Odoo's documented External API. Both are solid; we are just framing it honestly.

Use cases: what can an Odoo MCP server do in practice?

1. Creating & tracking orders “Create a sales order for customer X with these items.” The agent creates a sale.order, takes over partner, prices and tax rates from the master data and reports the status back: draft, confirmed or delivered.
2. Invoices & open items The agent reads account.move, answers “Which invoices are more than 30 days overdue?” and creates an invoice draft from a confirmed order; posting only after your approval.
3. Product & price maintenance Create new items in product.product, maintain prices and descriptions, check variants. The agent flags missing fields such as unit, tax class or supplier code before anything lands wrongly in the shop.
4. Contact & master data (res.partner) Reconcile customers and suppliers in res.partner, detect duplicates, complete VAT ID and address. Sync with a CRM so master data is not maintained twice.
5. Inventory & availability “How many units of item Y are available in the Hamburg warehouse?” The agent queries stock.quant or the stock models and can prepare a procurement notice as a draft when a threshold is undercut.
6. Reporting across modules Revenue by sales team, top customers in the quarter, goods receipts of the week: the agent pulls the figures together from several Odoo models, instead of you exporting views and marrying them in Excel.

How the integration works

Architecture overview for Odoo:

AI agent ↔ MCP server (your EU host) ↔ Odoo External API (XML-RPC / JSON-RPC / JSON-2)

Concrete technical steps:

• Transport & version: For Odoo 16-18 we use XML-RPC or JSON-RPC via the endpoints /xmlrpc/2 and /jsonrpc; from Odoo 19, the newer External JSON-2 API with an API key as bearer token. According to the Odoo documentation, /xmlrpc, /xmlrpc/2 and /jsonrpc are due to be removed with Odoo 22 (autumn 2028) and Odoo Online 21.1 (winter 2027) (Odoo External JSON-2 API). We therefore build new integrations on JSON-2, and older versions with a migration plan.
• Check the plan: According to the same documentation, data access via the External API is only available in Odoo's Custom plans, not in One App Free or Standard. We clarify that before the first line of code.
• Authentication: Instead of a plain-text password, we generate a dedicated Odoo API key (Settings → Users → API Keys) and store it encrypted in the MCP server. The technical user gets only the rights it really needs.
• Tool mapping: We define MCP tools such as search_records, create_sale_order, get_invoice, check_stock, each with a JSON schema, validated inputs and clear error messages. In the background they run through Odoo's execute_kw against the respective models.
• Permissions per model: We define granularly which Odoo models the agent may read or write, with no blanket admin access across the whole database.
• Read vs. write: By default, the agent reads or creates drafts. Confirming, posting or sending requires human approval, configurable per use case.
• Protective mechanics: Rate limiting and caching prevent an agent from overrunning the Odoo instance with requests; every action is logged with timestamp, user and diff.

GDPR, hosting & security

With Odoo, the hosting model mainly decides the GDPR situation: on-premise and Odoo.sh can be operated in an EU region, while with Odoo Online you should check the region and the DPA (data processing agreement) with Odoo. On this basis we harden the MCP connection as follows:

• EU hosting of the MCP server (e.g. Hetzner or Azure region Germany West Central in Frankfurt), no US edge in between.
• API key instead of password, stored encrypted, with a rotation and revocation strategy and a technical user with minimal rights.
• DPA with the AI provider with EU data residency: Claude via AWS Bedrock (EU region Frankfurt) or Google Vertex AI (EU), OpenAI Enterprise EU or Azure OpenAI in an EU region.
• Data minimisation: The agent reaches only the models and fields that the respective use case needs, with no blanket access to the whole ERP database.
• Multi-company separation: In Odoo setups with several companies (multi-company), the separation is preserved through the agent as well.

What does it cost?

We work in clearly delimited use cases instead of a big-bang project:

• First a technical check of your Odoo version, modules and depth of customisation, which gives the realistic effort.
• First use cases as a read-only start, typically order or invoice queries, then write actions with approval. We set the timeline after the AI Check.
• Clear scope per use case, no hourly-rate roulette. Scaling afterwards in small increments.

We name a concrete figure after the check, depending on volume, number of connected modules and the write permissions you want.

Frequently asked questions

Is there an official MCP server for Odoo?

We are not aware of a standard server published by Anthropic or Odoo SA. There are community solutions, though, including a widely used Python package on PyPI and a module in the Odoo App Store. Where one of them fits, we use it; otherwise we build cleanly on Odoo's documented External API.

Which interface does the MCP server use to talk to Odoo?

Odoo's External API. For Odoo 16-18 that is XML-RPC and JSON-RPC; from Odoo 19, the External JSON-2 API with an API key as bearer token. According to Odoo, the old endpoints /xmlrpc, /xmlrpc/2 and /jsonrpc are due to be removed with Odoo 22 (autumn 2028) and Odoo Online 21.1 (winter 2027), so we build new integrations on JSON-2.

Which Odoo objects can the agent address?

Practically any model that your modules provide: sale.order (orders), account.move (invoices), product.product (products), res.partner (contacts), the stock models (inventory) and more, limited by the rights we give the technical user.

Does this work equally with Odoo Online and self-hosted?

Technically yes, but not on every plan. According to the Odoo documentation, data access via the External API is only available on the Custom plans, not on One App Free or Standard. We check this before the start. For privacy-sensitive cases we recommend self-hosting or Odoo.sh in an EU region.

Can Odoo + MCP be operated in a data-protection-compliant way?

Yes, if the architecture is right: Odoo in an EU region, an EU-hosted MCP server, an AI provider with EU data residency and a DPA, and granular permissions per model. We check this for your specific setup in the check.

Which AI models work?

All MCP-compatible models: Anthropic Claude, OpenAI GPT, Google Gemini, Microsoft Copilot. A change of model requires no new integration; the MCP server stays the same.