AGENTIC WORKFLOWS FOR ODOO 19 · IMPLEMENTATION BY BLOOPARK
An AI agent that takes over a real process – within your rules
Connecting AI to Odoo is solved – for this we have our free MCP module. The actual work begins afterwards: Which process should an agent take over, what can it touch, where does a human still decide, and how do you prove that afterwards? That is exactly what we are building.
Discuss use case How we proceed
From €4,900 net for a use case · up to 8 person days · the basis is the free MCP module
Official Odoo partner · MCP productive in-house in use · View references
In summary
What is an Agentic Workflow?
A chat that can read Odoo answers questions. An Agentic Workflow takes over a work step: It collects what it needs, prepares a decision, and writes – if allowed – the result back to Odoo. The difference is not the model, but that a process is defined, limited, and traceable.
What does it cost?
A clearly defined use case starts at €4,900 net and is for eight person days designed. Multiple workflows, multiple agents or other systems are an individual implementation – offer after discovery.
Important for differentiation
We do not promise autonomous automation. An agent works within the models, fields and datasets that you release, and at defined points a human continues to decide. This is not a limitation, but the condition for something like this to run productively in an ERP.
Why at all
The most expensive part of a process is rarely the decision.
It is the gathering beforehand. Someone opens four Odoo views, reads a history, checks an inventory, formulates a summary – and then makes a decision that requires experience in two minutes. This preparatory work can be delegated to an agent without relinquishing the decision itself. That is why every one of our projects begins with the question of where in your process the preparation ends and the judgement begins.
The foundation
Governance comes first, then the agent
Every workflow runs through our MCP module: nine permission gates, approval before any write access, and an audit history linked by SHA-256.
The module itself is free of charge.
What companies start with
A good first use case is recurring, well-defined and has a clear point where a human agrees. These four meet that – your own case may be different.
Pre-qualifying leads
The agent reads incoming leads, supplements missing information from existing Odoo data, suggests evaluation and next steps. Assigned and passed on after confirmation.
Prepare conversations
Before an appointment, the agent compiles what is scattered in CRM, orders, and projects, and provides a brief preparation instead of four open views.
Internal information
Employees ask in natural language about orders, deliveries, or project statuses – limited to the models and datasets that their role is allowed to see.
Prepare records
Recurring setup and maintenance tasks are prepared and presented as a preview. Writing only occurs after approval, with before and after values in the log.
Which Odoo areas are relevant are determined by your approvals, not the module: it is model-agnostic. In practice, these are usually CRM, sales, projects, helpdesk, and inventory. If a process extends beyond Odoo, we connect the necessary systems via custom interfaces.
How an agent accesses Odoo – and what limits them
Four layers. Each can only restrict what the layer above allows.
1 · MCP Client
Claude, ChatGPT, Cursor or an automated service – an application that you choose and operate. It makes the request and has no rights of its own.
2 · Udo MCP
The endpoint in your Odoo. Checks emergency stop, agent status, rate limit, model and operation release, field policy and record filter.
3 · Release
Writing steps go through preview and confirmation. The token is unique and tied to a hash of the exact values.
4 · Odoo itself
Finally, the native access rights of the stored user apply. The MCP level can only revoke rights, never add them.
Every call – including every rejected one, with reason and rejecting gate – ends up in a SHA-256 linked, append-only history. Changing and deleting are locked at the ORM level. This is visible to manipulation against application-level interventions, not against direct database access. Technical details on the MCP product page.
This is how a project proceeds
Eight person-days, divided into four sections. Each ends with something you can view and approve – not with a status report.
01
Discovery
Half a day with the people who are doing the process today. We choose exactly one use case and write down how success will be measured. If none of this fits, we say so here – not after eight days.
02
Set boundaries
Which models, fields, and datasets the agent is allowed to see, what they are allowed to write, and at which point a human confirms. The result is an authorisation concept that is also readable without us.
03
Build and test
Instructions, tool integration, and release points are implemented and tested on real data – including cases where the agent is not allowed to do anything.
04
Acceptance and handover
Joint run-through, documentation, briefing of those involved. After that, 30 days of follow-up for corrections to the agreed scope.
What is not included in the starter package
So that "from €4,900" remains a reliable figure and not a negotiation position: Not included are a second use case, the integration of systems outside of Odoo, custom Odoo modules or MCP tools, cleaning of existing data quality, operation and monitoring beyond the 30 days, as well as DSFA and AVV documentation. These are not exclusions against you – they are the points where other projects quietly become expensive. We can do all of this, as individual implementations or at an hourly rate of €171.50, always after prior agreement.
What it costs
One use case for a fixed-price entry, everything larger after discovery. The software underneath remains free in any case.
Prerequisite
MCP module
The technical foundation. You can install it yourself – or we can take care of it.
0 €
free · setup from 1,490 €
- Complete module, nothing held back
- Nine governance gates and audit chain
- Setup by your team or by us
The entry point
Agentic workflow
A process, fully implemented, accepted and handed over.
from 4,900 €
net · up to 8 person days
- Discovery and selection of a use case
- Authorization and approval concept
- Implementation, testing on real data, acceptance
- Documentation and training
- 30 days of follow-up support
Anything larger
Custom implementation
Multiple agents, multiple systems, corporate requirements.
On request
Offer after discovery
- Multiple workflows and agents
- Integration of additional systems
- Own MCP tools and Odoo modules
- Operation, monitoring, further development
All prices net, plus statutory VAT. The entry price assumes a functional Odoo-19 instance, an accessible MCP endpoint, and responsive subject matter experts.
Support
Not there yet? Questions are still welcome.
You do not need to commission a project to talk to us. If you try the free MCP module and get stuck somewhere – installation, permissions, a client that won't connect – use the helpdesk. And if you are unsure whether your case is suitable for an agent: That is a good question for an initial conversation, not for a quote.
Helpdesk
Technical questions about the module, regardless of whether you are a customer.
Support anfragenOr directly: +49 391 5630690
FAQ: Agentic Workflows in Odoo
Frequently asked questions about scope, limits, costs, and responsibilities.
The connection is the first step and costs us nothing – the MCP module is free. After that, an assistant can answer questions about your data. An agentic workflow goes further: it takes over a defined process, with set boundaries, defined approval points, and a protocol. The effort is not in the technology, but in these specifications.
Only where you explicitly allow it – and writing steps generally go through preview and human confirmation. In practice, the desired pattern is almost always: The agent prepares, a human decides. We do not sell autonomous automation, and we would also advise against it in an ERP.
Three things apply. He can only access models, fields, and records that are released – everything else is rejected and logged. Writing steps require a confirmation, and the confirmation token is tied to a hash of the exact values you have seen. And a global emergency stop immediately halts all MCP requests without restarting Odoo.
Through the audit history. Users, agent, company, time, tool, model, records, requested and masked fields, status, and duration are recorded; in case of rejection, the rejecting gate with reason is additionally noted; for changes, the before and after values are documented. The entries are linked via SHA-256 and are neither changeable nor deletable at the ORM level. This is tamper-evident, not tamper-proof – against application-level interventions, not against direct database access. Every provider should mention this difference to you.
For exactly one clearly defined process: yes – the package is designed for eight person-days. What does not fit is explicitly stated above as not included. If it becomes clear during discovery that your case is larger, we will inform you before commissioning and make a separate offer. A project that is calculated too small helps no one.
Only what the AI client requests: the released fields of the requested records, to the model provider you choose. The processing otherwise takes place in your own Odoo database; bloopark receives no data. How much can actually be transferred is determined by your field policies – this is data minimisation according to Art. 5 GDPR, technically enforced rather than promised. You conclude the data processing agreement with your model provider.
We will decide this with you based on requirements, data protection situation, and costs – the MCP module is model-independent and speaks standard MCP. In practice, Claude, ChatGPT, and Gemini are the usual candidates. The licensing or usage costs of your model provider are not part of our offer.
The transparency obligation under Art. 50 EU AI Act is directed at the operator, not at the model provider. Those who allow AI agents to access business data are themselves responsible for proof. This is a good reason to set up access to be loggable from the start rather than retrofitting later. This is not a complete legal consultation – that is the responsibility of your legal advisor.
Which process would you like to hand over?
If you have an answer to this question, the initial conversation is short and specific. If not, that is exactly the first item on the agenda. Before that, you can try out the free MCP module at any time.
Discuss use case Download module for free
Questions in advance? Helpdesk · Contact · +49 391 5630690