MCP integration
Connect AI to your systems — without migrations.
The Model Context Protocol (MCP) is the open standard that lets artificial intelligence models query data and take actions in the applications you already use. Datapoint designs, builds, hosts and maintains the MCP servers that make it safe.
- Hosted in the EU
- Per-user permissions
- Full audit trail
For decision-makers
What MCP is, in plain language.
Think of a universal socket. Instead of building a different integration for every AI model and every system, MCP defines a single way of connecting the two.
- Your systems stay where they are. The MCP server is a thin layer on top of them.
- AI only does what is authorised. Every tool has permissions, limits and logging.
- Choose any model. Claude, GPT, Gemini or open-weight models use the same connector.
- Every connector is reusable. What you connect today for one assistant serves another tomorrow.
An open standard
Published in 2024 and adopted by the leading AI model providers.
Replaces nothing
Works on top of existing APIs, databases and exports.
Governed
Permissions, auditing and approvals in a single control point.
No lock-in
Change AI models without rebuilding integrations.
How it works
From request to answer, in four steps.
Choose the explanation for decision-makers or for technical teams.
Who asks
EU perimeter
AI model
Claude, GPT, Gemini or open-weight models
MCP servers
- Per-user permissions
- Audit trail
- Human approval
Your systems
For technical teams
The three building blocks of an MCP server.
Each server exposes well-defined capabilities, described by schemas the model understands.
-
Tools
Actions the model can invoke, such as
cases.searchorinvoices.post. Arguments are validated by JSON schema; write tools require confirmation. -
Resources
Data the model can read as context: documents, records, reference tables — always filtered by the user's permissions.
-
Prompts
Predefined instructions and flows for recurring tasks, ensuring answers consistent with the organisation's rules.
What we connect
If it has data or an API, we can connect it.
Examples of systems we connect to AI models through MCP servers.
-
Databases
PostgreSQL, SQL Server, Oracle, MySQL — with parameterised queries and dedicated views.
-
ERP and finance
Invoicing, purchasing, accounting and stock, through APIs or integration files.
-
Document management
Records, workflow and administrative cases, with document search and reading.
-
CRM and helpdesk
Customers, tickets and contact history, for informed support.
-
HR and payroll
Directory, leave, training and internal procedures, with restricted access.
-
Calendars and bookings
Availability, bookings and rescheduling, with human confirmation.
-
Email and Microsoft 365
Shared mailboxes, calendars, Teams and SharePoint, with Entra ID authentication.
-
Legacy applications
Older systems without a modern API, through databases, files or dedicated adapters.
Benefits
Why MCP changes how organisations adopt AI.
-
Protects your investment
The systems you have already invested in become the engine of AI, instead of being replaced.
-
Always-current data
AI queries the source at the moment of the question. There are no copies to keep in sync.
-
Supplier independence
The same MCP server serves different models. Choosing a model is no longer irreversible.
-
Central control
Permissions, limits, approvals and auditing defined once, applied to every assistant.
-
Fast return
A well-chosen first connector delivers impact in weeks, not years.
-
Scales across the organisation
Every new connector becomes available to every department and assistant.
Security and governance
AI only sees what the person could already see.
Security is not an add-on: it is the reason a well-designed MCP server exists.
- Least privilege: each tool accesses only what is strictly necessary.
- Read-only by default: write tools are explicit and require human approval.
- User identity: OAuth / Microsoft Entra ID authentication; permissions are the person's, not the AI's.
- Auditing: every call is logged — who, when, which tool, with which arguments and result.
- Personal data: masking and pseudonymisation before anything reaches the model.
- Prompt-injection defences: argument validation, allow-lists and external content treated as data.
{
"name": "cases.route",
"description": "Routes a case to a service.",
"inputSchema": {
"type": "object",
"properties": {
"ref": { "type": "string", "pattern": "^REQ-[0-9]{4}$" },
"service": { "enum": ["Planning", "Mobility", "Front office"] }
},
"required": ["ref", "service"]
},
"annotations": {
"readOnlyHint": false,
"destructiveHint": false,
"idempotentHint": true
},
"_meta": {
"datapoint/approval": "required",
"datapoint/audit": "cases"
}
}How we deliver
From the first connector to continuous operation.
-
Design
We identify processes, systems, personal data and approval points, and define tools and their limits.
-
Build
We build the MCP servers, with automated tests, schema validation and documentation.
-
Host
We go live in the European Union, through our partners, or on your infrastructure.
-
Maintain
Monitoring, security updates, evolving tools and new connectors.
We use what we recommend
MCP is part of our everyday work.
At Datapoint, our own project and ticket management platform is connected to AI assistants through MCP — read-only and with each employee's permissions.
- Ticket and project status summaries in seconds.
- Search across technical history and internal documentation.
- Time and effort reports without manual exports.
Search
Find tickets, decisions and technical documentation in a single question.
Reporting
Effort and deadline indicators generated from real data.
Demonstration
Watch an MCP server working with an ERP.
Reading invoices, matching purchase orders and pre-posting — with approval before writing.
Invoice processing — Distributor (simulation)
Processed in the EUPedro has 120 supplier invoices in the inbox this morning.
MCP calls
0- ERP
Simulation with fictitious data. No real system is contacted.
Frequently asked questions
About MCP integration.
Is MCP a proprietary technology?
No. The Model Context Protocol is an open standard. The solutions we build on it are licensed by Datapoint and tailored to each client.
Do we need to change systems?
No. MCP servers connect to existing systems through their APIs, databases or exports.
Which AI models are compatible?
The main commercial models (such as Claude, GPT or Gemini) and open-weight models hosted in the EU. We choose according to the task and the sensitivity of the data.
Who has access to the data?
Only the people who already had access. The MCP server uses the identity of the person asking and logs every access.
What if the system has no API?
We connect through the database, files or dedicated adapters. The initial assessment identifies the best route.
Which system would you like to connect first?
Tell us. We assess the connection options and propose a first connector with measurable impact.