UDO – MCP SERVER FOR ODOO · FREE MODULE · DEVELOPED BY BLOOPARK
Odoo MCP Server: Work faster with AI – You retain control and proof
Udo – MCP Server for Odoo connects AI assistants like Claude and ChatGPT with your Odoo: Analyses, research and routine processes run in dialogue instead of click paths. You set the framework – write access only after human approval, every access logged in a tamper-evident chain. The module is free. You only pay if setup, integration or a real AI workflow from Odoo partner bloopark is required.
Download module for free Help with the setup?
Odoo 19 · Open-source module, free · Installation by your team or by bloopark · Odoo Online (SaaS) does not allow custom modules
Official Odoo partner · actively used in-house · 621 automated tests · View references
In summary
How does it work?
An Odoo module provides an endpoint: /mcp, JSON-RPC 2.0 over Streamable HTTP. The connecting MCP client logs in via OAuth 2.1 with PKCE, you confirm access on an Odoo consent screen. Each call then passes through nine permission gates.
What does it cost?
Das Modul kostet nichts. Wer Odoo-Module selbst installiert, lädt es und legt los. Bezahlt wird die Arbeit drumherum: Einrichtung & Integration für 1.490 €, ein Agentic Workflow ab 4.900 €, oder eine vollständig individuelle Umsetzung auf Anfrage.
Important for differentiation
Currently Odoo 19. Odoo Online (SaaS) does not allow custom modules. The module is always complete – there is no paid upgrade and no function that is behind a purchase. Payment is made solely for working time.
Why Governance
The bottleneck is not the connection. It is the approval.
Connecting AI to an ERP is now easy. Until now, there were two ways to do this: an unrestricted API key, meaning all or nothing and without a trace, or custom integration code for each use case. Neither answers the question that management, auditing, and data protection ask: What is the AI actually allowed to do here? This module answers that question – completely, and without charging for it.
In-house operation
Productive in our own Odoo 19
bloopark operates the MCP server in-house. The foundation of this page was also created through the MCP server itself: An AI agent read 29 approved models for this.
Quality signal: 621 automated tests over 30 test files
This is how Udo MCP works in practice
The video shows the complete process: installation of the module, configuration of role and agent, a brief explanation of the setup, and the first practical interaction via an MCP client – including example prompt.
What you see in the video
- Installation – search for and activate the module in Apps
- Configuration – authorise yourself, turn on MCP, copy /mcp address
- Setup explained – how roles, agents and permissions interact
- First prompt – connect MCP client and ask the first question
Example prompt in the video
Show me the open opportunities over 10,000 euros.
After that: how write access is granted and what the documentation covers.
Each connection is granted individually
Before an MCP client accesses Odoo for the first time, Odoo displays a permission dialog. It names the client, the logged-in user, and the company – and lists what the agent is allowed to do afterwards. Agreeing or declining is decided by a human.
Read business data
Contacts, orders, invoices and projects – within the framework of the role and the rights of the registered Odoo user.
Create and modify
Explicitly marked in the dialog as elevated permission and solely under your governance rules and the permission workflow.
Maintain connection
The connection between sessions remains via refresh tokens. This point is also marked in the dialog as elevated permission.
Connect read-only
A tick in the dialog completely removes the write permission for this connection: read yes, create, modify or delete no – for the entire session, even after its renewal.
Each action subsequently lands in the audit log. You can revoke a client's access at any time under MCP AI Governance → Configuration → Agents; the global emergency stop in the settings also stops all MCP requests in one go.
Write access only after approval: the three steps
Every change to Odoo data goes through preview, approval, and execution. There is no direct write path.
1 · Preview
The agent calls up the preview. A field-level diff and a one-time HMAC token are returned.
- Field diff before the change: old value, new value
- One-time token, valid for ten minutes
- Bound to agent, model, operation, and record
- Bound to a hash of the exact values
2 · Approval Core
A person confirms the shown diff. If the values change after the preview, the token is invalid.
- The confirmed diff is the executed diff
- No silent changes between preview and execution
- The decision lies with the person, not the model
3 · Execution
Only now does Odoo write. Before and after values go into the audit chain.
- Execution only with a valid token
- Before and after values are logged
- Rejected accesses are logged with reason
Nine gates: what the AI is allowed to do is decided by Odoo
Every request goes through nine ordered checks: emergency stop, agent and credential status, rate limit, model allowlist, operation allowlist, scope, field policy (also over nested relationships), record filter, and finally the native Odoo access rights. The audit entry is written as fail-closed.
Gate 9 is the crucial one: It checks the native Odoo access rights as the linked user. Therefore, the MCP level can only revoke rights, never add them. No second rights system is created that runs alongside your Odoo groups and record rules.
- No rights extension possible, the native Odoo ACL decides last
- Passwords, TOTP secrets, API keys and salary fields are fundamentally masked
- sudo, with_user, execute_kw and raw create, write, unlink are permanently blocked
- A global kill switch stops any AI access without restarting Odoo
Models
Allowlist per role
Only explicitly approved models are visible at all. In our live operation, there are 29.
Fields
Allow, block, mask
Field policies also apply across nested relationships.
Records
Domain filter per agent
A filter limits access to exactly the records that are meant.
Tenants
Company boundary and limits
Agents are limited to their companies. Additionally, there is a rate limit per role – as a security boundary, not as a pricing tier.
Provable: Audit chain and emergency stop
Anyone allowing AI on business data must be able to prove what has happened. This is not an afterthought here, but a separate gate.
Hash chain
Each entry hashes the previous one with SHA-256. Subsequent changes break the chain and make it visible.
Append only
The protocol is append-only. Modifying and deleting are locked at the ORM level, not just by permission.
Automatic check
A daily job checks the integrity of the chain, an hourly one reports anomalies, a daily one enforces retention.
What it contains
User, agent, company, timestamp, tool, model, records, requested and masked fields, status, duration. And in case of rejection: which gate was rejected and why.
Kill Switch
A switch stops all AI access immediately, without restarting Odoo. Gate 0, before anything else.
Extensible
Through a dynamic tool registry, add-on modules bring their own MCP tools without changing the core.
What an AI agent can read and write in Odoo
The MCP is model-agnostic: access is granted per role, not per app. Eighteen tools are available, along with a registry through which add-on modules can contribute their own tools.
Read
- Search with domain filters, sorting, pagination
- Read individual records
- Grouped evaluations
- Model and field introspection, access diagnosis
Write
- Create, modify, delete, each in two stages
- Post chatter messages
- Call released business methods (Beta)
- Release multiple changes as a bundle (Beta)
Example: Read
Request: Show me the open opportunities over 10,000 euros.
The MCP searches on crm.lead with domain filter and returns the filtered records, masked locked fields, access logged.
Example: Write
Request: Set the price of product X to 49 euros.
The preview shows list_price 42.00 to 49.00 and issues a token. You confirm, the execution writes, before and after values land in the audit chain.
What you release is released. In live operation at bloopark, 29 models are open, including contacts, leads, orders, products, invoices, projects, and calendars. For systems beyond Odoo, we additionally connect in consulting custom interfaces, and for the accounting side, everything that is needed for a clean Odoo accounting is applicable.
Set up yourself: source code, instructions, video
The free module is documented, not supported. Source code, installation instructions, and product video are public – your team handles the setup. The initial connection is completed in about five minutes according to the installation instructions and runs entirely in the Odoo interface; you can see the steps in the video. Field policies over nested relationships and record filters require knowledge of the Odoo models – at this boundary, consulting is worthwhile.
01
Activate module
In Odoo, search for the module under Apps and click Activate. Odoo installs it and reloads. The predefined MCP roles as well as the background jobs for logging and retention are created.
02
Authorize yourself
Set the access rights MCP AI Governance to MCP Administrator in your own user. Without this role, the menu and settings remain hidden.
03
Switch on MCP
In the settings under MCP AI Governance, check the box for MCP Platform Enabled and copy the /mcp address displayed below.
04
Connect client
Enter the address in the MCP client as a connector, log in to Odoo, and confirm the connection in the sharing dialog – read-only if desired.
The module is free. Payment is for the work.
Four ways, depending on how far you want to go yourself. The functionality of the module is identical in all four – it is never the difference.
You do it yourself
Free MCP module
Download, install, configure. Right for teams that deploy Odoo modules themselves.
0 €
permanently free · Open Source
- Complete functionality, nothing withheld
- Read and write tools with sharing workflow
- Nine governance gates, field masking, audit chain, emergency stop
- All supplied roles
- Documentation and source code on GitHub
- Setup by your team · no support
We set it up
MCP Setup & Integration
We install and configure the MCP module in your Odoo instance, set up authentication and permissions, and review the integration together with you.
Fixed price · plus VAT
- Installation and activation in your instance
- Role and rights configuration according to documented role list
- Setup of the secure MCP connection including authentication via OAuth 2.1
- Field and record sharing for the agreed scope
- Connection and sharing test, jointly approved
- Handover with customised documentation · 60-minute briefing
We build a workflow
Agentic workflow
A AI agent takes over a real business process – with limits, approvals, and protocols.
from €4,900
a use case · up to 8 person-days
- Discovery and selection of exactly one use case
- Workflow architecture and connection of the necessary Odoo models
- Instruction and permission design
- Defined human approval points
- Implementation, testing and joint acceptance
- Documentation, handover, 30 days post-support
Discuss workflow
We design the solution
Custom implementation
Multiple agents, multiple systems, proprietary tools, corporate requirements.
On request
Offer after discovery
- Custom MCP architecture and own MCP tools
- Multiple workflows and agents
- Integration of additional systems beyond Odoo
- Approval and governance concepts for corporate structures
- Own Odoo modules and interfaces
- Operation, monitoring and further development
What exactly you pay for
None of these stages unlocks a function. The module is the same in all four cases, complete and free of charge. Only working time is charged: configuration, permission concept, connection, automation, testing and handover. Those who start with the free module lose nothing – we start where you are. Services outside the agreed scope are charged transparently at an hourly rate of €171.50, after prior agreement.
Which level suits you?
The question is not which features you need – they are always all included. The question is how much work you take on yourself.
| Free module | Setup & Integration | Agentic workflow | Individual | |
|---|---|---|---|---|
| The statement that applies to you | "We will install it ourselves." | "Please make the MCP operational for us." | "We want AI to take over a process." | "This involves multiple systems and departments." |
| Complete module, all tools and gates | included | included | included | included |
| Installation and configuration | Your team | bloopark | bloopark | bloopark |
| Permission concept and field policies | Your team | according to role list | included | included |
| Setting up the MCP connection | Your team | included | included | included |
| Defined business process as AI workflow | – | – | a use case | multiple |
| Instruction and approval design | – | – | included | included |
| Integration of additional systems | – | – | – | included |
| Multiple agents and workflows | – | – | – | included |
| Own MCP tools and Odoo modules | – | – | – | included |
| Aftercare | Community & documentation | Handover briefing | 30 days | by arrangement |
| Price | 0 € | 1,490 € | from 4,900 € | On request |
All prices are net, plus statutory VAT. Setup & integration require an Odoo-19 instance, an MCP-capable tool on your side, and standard hosting; on-premise special cases, SSO, network shares on the customer side, permission mapping for custom modules, as well as DSFA and AVV documentation are not included and will be offered separately. "Beta" means: functional, but not yet widely tested.
Odoo AI interface: what matters in the evaluation
The connection itself has now become standard. It becomes interesting at three points: where your data goes, what happens when an agent is manipulated, and whether you can later prove what has happened. These three questions are exactly what management, auditing, and data protection ask in German SME projects.
Where are our data?
Everything remains in your own Odoo database. bloopark does not receive any data. Your house only sends what the MCP client requests, namely the released fields of the requested records, and that goes to the model provider you have chosen. That is exactly why there are field policies: The release at the field level is data minimisation according to Art. 5 GDPR, technically enforced rather than organisationally promised. The protocol supports accountability according to Art. 5 para. 2 and helps with data subject requests. You conclude the data processing agreement in self-operation with your model provider, not with us.
Why is the module free?
Because access to a protocol is not a product. In the Odoo ecosystem, there are now several free MCP modules; charging for the pure connection would be a bet against the market. What costs money is not the connection, but everything that makes it responsible: a permission concept that withstands scrutiny, field policies over nested relationships, record filters, and the decision about which processes an agent is allowed to touch at all. We sell this work – not the download. This does not change the scope of functionality: There is no stripped-down version and no function that is only unlocked in consulting.
What about Prompt Injection?
To be honest: Prompt Injection cannot be patched away. The answer must therefore be architectural, not cosmetic. By default, only read access, human approval for every write access, field allowlists, record filters, rate limits, and a log in which an outlier becomes visible. An assumed agent can then only read what the linked user was allowed to read anyway. Nothing more. This is not complete immunity, but it is a verifiable limitation.
And what if Odoo brings that itself?
A dedicated MCP server is expected for Odoo 20. We will evaluate this after the release, not after the announcement. Three questions will then decide: whether it is also available in the Community Edition and in older versions, whether write accesses require human approval, and whether there is a searchable log including approval history and emergency stop. Integration will become a commodity – that’s why we charge nothing for it. The question of whether you can prove what the AI has done remains.
Does the EU AI Act concern us?
The transparency obligation under Art. 50 EU AI Act has been in effect since 2 August 2026 and is directed at the operator, not the model provider. Anyone allowing AI agents access to business data is therefore responsible for proof themselves. This is a good reason to set up access logging from the start rather than retrofitting later. How we approach the topic of automation and AI otherwise is shown on our page about autonomous marketing automation with AI.
Can you install Odoo modules yourself? Then download the free module and get started. If configuration, integration, or a Agentic workflow should be with us, schedule an appointment. How we have implemented comparable Odoo projects is shown by our customer references.
support
Stuck? Ask us.
The helpdesk is also open to users of the free module. Installation issues, questions about configuration, a client that won't connect, a permission that doesn't behave as expected: Describe the case, and you will receive a response from someone who built the module.
Is it not about an incident, but about setup, integration or an agentic workflow? For that, the initial conversation is the shorter way.
Helpdesk
Please state the Odoo version, the MCP client used, and the point where it is stuck – this saves a follow-up question.
Support requestsPrefer to call? +49 391 5630690
FAQ: Odoo MCP Server
Frequently asked questions about packages, pricing, support, documentation, and setup.
Nothing. The MCP module is free and complete: read and write tools with a two-step approval workflow, all nine governance gates, field masking, the tamper-evident audit trail, the emergency stop, and all supplied roles. There is no stripped-down version and no feature that is behind a payment. Money only costs working time: setup, integration, an agentic workflow, or a custom implementation.
Through the Odoo Apps Store, free of charge. You can find the source code on GitHub. It is installed in your own Odoo instance; nothing runs on our servers.
Yes. Our Helpdesk steht Ihnen offen, auch ohne Vertrag – wir beantworten Fragen zu Installation, Konfiguration und Client-Anbindung. Was der Helpdesk nicht ist: ein Ersatz für ein Berechtigungskonzept oder eine Projektbegleitung. Sobald es darum geht, lohnt sich Setup & Integration.
For the work that the module does not take over. Which models an agent is allowed to see, which fields remain masked, which dataset filter applies, where a human must approve, and how this can be substantiated in relation to auditing and data protection – these are decisions, not settings. Additionally, everything that goes beyond an Odoo instance: other systems, multiple agents, custom MCP tools.
Yes, and that is the usual way. Since it is always the same module, nothing changes in your installation. Roles, agents, and approvals are Odoo records in your own database and remain intact. We start where you are.
The initial connection takes about five minutes according to the installation guide: activate the module in Apps, assign yourself the role of MCP Administrator, enable MCP in the settings, copy the /mcp address into the MCP client, and release the connection. The product video shows exactly this path up to the first prompt. You should plan write access with field policies and dataset filters for one to two days, spread over two sessions, so that testing can be done between configuration and approval.
Yes. The approval dialog when connecting offers the option Connect read-only. It withdraws the write permission from this connection: The agent can read, but cannot create, modify or delete – for the entire session, even after its renewal. Regardless, reading remains the default state: Writing tools only take effect when the role releases them and a person confirms the specific change.
Under MCP AI Governance → Configuration → Agents, you can see every connected client with type, role, associated Odoo user, and last activity. An agent can be deactivated there; access ends as a result. Additionally, the global emergency stop in the settings halts all MCP requests at once.
Installation of the module, activation in the settings, creation of roles and agents, setting up MCP connections via OAuth or API key, as well as the configuration of model, operation, and field permissions. Field policies over nested relationships and record filters require knowledge of the Odoo models – that is the point where consulting becomes meaningful.
Yes, and that is the core of the architecture. Gate 9 checks the native Odoo access rights as the linked user. The MCP layer can only restrict rights, never expand them. No second rights system is created that runs alongside your groups and record rules.
The processing takes place in your own Odoo database, bloopark does not receive any data. Your house only sends what the MCP client requests: the shared fields of the requested records, to the model provider of your choice. Where this goes geographically is determined by your choice of provider, and how much can be transferred at all is determined by your field policies. You conclude the data processing agreement with your model provider, not with us.
Through a protocol that can only be appended and is hashed into a chain via SHA-256. Subsequent changes break the chain and make them visible. Captured are, among other things, user, agent, company, time, tool, model, records, requested and masked fields, status, duration, in case of rejection the rejecting gate with reason and in case of changes the before and after values. Three background jobs check integrity, anomalies, and retention.
Currently Odoo 19, built and tested. Both Community and Enterprise are supported, as well as On-Premise and Odoo.sh. Odoo Online (SaaS) does not allow custom modules and is therefore not possible. Older Odoo versions are not supported today.
Basically, any MCP-capable client. The module speaks standard MCP and supports the protocol versions 2025-06-18, 2025-03-26, and 2024-11-05, which are negotiated during the connection setup. In practice, among others, Claude, ChatGPT, Cursor, and Google Gemini are used; interactive clients connect via OAuth 2.1, automations via an API key.
API keys are stored salted and hashed, with only a display prefix remaining in plain text. You will see the key itself exactly once, at creation. A rotation keeps the old hash valid until the end of a grace period. OAuth tokens are stored as a digest, authorisation codes are one-time and valid for ten minutes, refresh tokens rotate upon use. Passwords, TOTP secrets, API keys, OAuth tokens, and salary fields are fundamentally masked for each role, even without a configured policy.
This is decided by three questions that we will check after the release: Is it also available in the Community Edition and in older versions? Does it require human approval for write access? And is there a searchable, audit-proof log with approval history and emergency stop? The pure connection becomes a commodity, the traceability does not.
Download it. It costs nothing.
If your team installs Odoo modules themselves, get the module and start today – free, complete, with source code. If setup, integration, or a real AI workflow should be with us, we will discuss your needs.
Download module for free Schedule an appointment
Questions in advance? Helpdesk · Contact · +49 391 5630690