Connect your AI system to the LDB <i class="noprint fa fa-video-camera"></i>

Video

In this onboarding video, we show you step by step how to connect and use your AI system with the LDB via MCP:

Navigation

MCP is not enabled centrally, but selectively per query report and per import interface.

Read access:
Navigation bar → Settings → Query reports → edit the desired query report → Connections → MCP

Write/update access:
Navigation bar → Samples → Import interfaces → edit the desired import interface → Connections → MCP

You can find the templates under:
Navigation bar → Settings → System settings → AI integration → MCP templates

Note: The exact labels of the menu items may differ depending on the configuration of your LDB.

What is MCP?

With MCP, you simply ask your questions in your AI chat — the AI retrieves the answers directly from the released data of your LDB and, if you wish, updates it right there as well.

MCP (Model Context Protocol) is the common language in which the AI and the LDB communicate. So that the AI can use the LDB, you make individual tools available to it. A tool is a function you have released — a query report or an interface. For each tool, the AI automatically recognizes what it is for, which information it must provide, and in what form it receives the result.

There are two types of tools:

• Query reports are read-only tools: the AI uses them to retrieve released data from the LDB without changing anything.
• Import interfaces are write tools: the AI uses them to add new data or update existing data.

The AI therefore works exclusively through these clearly delimited tools — and not through direct, unrestricted access to the entire database. It can reach only what you have explicitly released.

Note: MCP is not an AI model itself. It merely connects AI applications such as ChatGPT or Claude with the functions and data of the LDB that you release for this purpose.

Overview

The Model Context Protocol (MCP) is an open standard that connects the LDB with AI systems such as Claude, ChatGPT, or Google Gemini. As a result, the AI can not only answer questions but also retrieve released data in a targeted way, record it in a structured form, and carry out defined work steps — controlled via the permission management that you already maintain anyway.

The MCP connection allows you to:

• Use a uniform AI connection based on an open standard — without a separate interface for each AI model
• Grant read access via released query reports
• Grant write or update access via released import interfaces
• Release individual objects selectively instead of unrestricted database access
• Apply user-specific access rights, including full traceability in the audit trail
• Automate recurring queries and clearly defined work steps

MCP calls run through your own AI system and do not consume any AI tokens of the LDB. A LDB license with AI integration is required.

Note: MCP is disabled by default. Every query report and every import interface must be deliberately released for MCP use, per user. This conservative default setting protects against unintended data access.

Enabling a query report for MCP

Query reports provide data for read access.

Activating MCP in the query report

• Open the edit view of the desired query report and switch to the Connections section.
• Enable the checkbox Publish Query report as an MCP tool.
• Assign a meaningful, descriptive tool name (e.g. sample_demo) so that the AI can correctly identify the tool.

Figure: Activating MCP in the query report

Further information: Create / Edit query reports

Adding a description / instructions for the AI

• Describe what the AI can filter by and which information is returned (e.g. the sample description and database ID based on the sample number).
• Name the relevant fields or variables. The more fields there are, the more detailed the description should be so that the AI knows which tool to use.

Figure: Description of the query report for the AI

Our recommendation: Make the description as precise as possible — which fields can be filtered by and which information is returned. This significantly improves the AI's hit rate.

Parameters and time period

Which search fields the AI can fill in when calling a query report follows automatically from the query report itself — you do not need to set anything up separately. If your query report has, for example, a search field "Sample number", MCP offers this field to the AI. The AI sees the field name you assigned in the query report.

If the user asks in the chat, e.g. "What is the description of the sample with sample number XY?", the AI recognizes that "XY" belongs in the "Sample number" search field, enters the value there, and calls the query report with it. As the result, it returns the corresponding description in the chat.

• If a search field has a preset value, the entry is optional; otherwise the AI must fill it in.
• If your query report works with a time period, the AI can additionally enter a start and an end date. Without an entry, the default time period set in the query report applies.
• Query reports without search fields and without a time period can be called by the AI without any further input.

Enabling an import interface for MCP

Import interfaces let the AI send data back to the LDB.

Activating MCP in the import interface

• Open the desired import interface and switch to the Connections section.
• Enable the checkbox Publish Import interface as an MCP tool and assign a meaningful tool name.

Figure: Activating MCP in the import interface

Adding a description / instructions for the AI

• Description: Describe when the AI should use this import tool and how the task should be solved. State, for example:
  • which information identifies the record (e.g. the database ID of the sample),
  • in which format date fields must be entered (e.g. DD.MM.YYYY),
  • how a status is set (using the key; the name and key of the status can be found using the tool status_search_details).
    Note: If nothing is entered here, the description or the name of the import interface is used.

Figure: Description of the import interface for the AI

Defining the import key and formats

In the General settings section, define the specifications for the import:

• Import key: Define what identifies the record (e.g. the LDB ID of the sample).
• Date format: Specify the date format so that no error occurs during the data import.

Note: For sensitive or write actions, you can use the associated MCP description to require the AI to obtain a confirmation before execution.

Input schema and import mapping

The MCP tool takes the input fields and data types from the interface's mapping. Fields other than those defined in the mapping are rejected. The records are then processed unchanged via the configured import interface.

Further information: Edit Attributes in Import Mapping

Retrieving the MCP server URL

At the bottom of the section where you activate the MCP connection, you will find the link to the MCP server of the LDB.

• After adding this single URL to your AI system, all query reports and import interfaces for which MCP is activated and for which you have the corresponding permission become available automatically.
• This link is identical for all users of your instance. The URL contains your instance name.

Figure: MCP server URL in the LDB

Note: Every MCP connection is tied to a user. The call runs through the AI account of the respective employee and is traceable accordingly in the audit trail of the LDB.

The following describes how to set up MCP in, for example, Claude and ChatGPT and connect your LDB application with the respective AI chat.

Setting up MCP in Claude

• In Claude, open the Connectors section and then Manage connectors. There you can also see which connectors are already activated.
• Click Add and assign a name.
• Enter the MCP server URL of your LDB and confirm. Claude then establishes the connection and redirects you to your LDB.
• Sign in with your LDB credentials and grant access.

The connection then appears in the tool list and can be used directly in the AI chat. For each connector, you can define whether its use is always allowed, prompted each time, or generally blocked.

Note: If you have created a new query report or import interface, refresh the tool list in Claude so that the new tool becomes available and you can access the data in the AI chat.

Setting up MCP in ChatGPT

• Open the Plugins section and, in the settings, the MCP section.
• Enter the name and the MCP server URL of your LDB and confirm. ChatGPT then establishes the connection and redirects you to your LDB.
• Select HTTP as the connection type. The STDIO (Standard IO) option is intended for locally connected programs and is not required here.

Since the connection is tied to your account, it is then available everywhere you use your account.

Our recommendation: For use in the accredited area, we recommend a paid AI variant for data protection reasons. The connection generally also works with the free variants.

Access rights and security

Through MCP, an external AI does not automatically gain access to all data of the LDB.

• Every query report and import interface must be released specifically for the user and for MCP use.
• In addition, the user-specific access rights apply: the AI sees only what the logged-in user is allowed to see.
• MFA and SSO also apply to access via MCP.
• All MCP calls are recorded at the respective query report or import interface, marked with "MCP", and assigned to the user — the same traceability as for an entry via the interface. Your existing release and validation steps remain unchanged.
• On the LDB side, the data transmitted via MCP is used exclusively to answer the respective request and is not used to train the models.
• How the AI assistant itself handles the transmitted data (e.g. use for training purposes) depends on the terms of the respective provider, the selected account type (free, Team, Enterprise, API), and your settings there.

Our recommendation: The principle of data minimization applies. Release only what the AI actually needs, and specifically exclude sensitive content from the released query reports.

Importing MCP templates

For getting started, ready-to-use MCP templates are available — no SQL knowledge is required, as the AI can handle the database queries. The templates cover, among other things, query reports for searching (sample orders, customers, prices, parameters, limit values) and import interfaces (measured values, orders, customers, samples).

• Open Navigation bar → Settings → System settings → AI integration → MCP templates. The group with the available templates is displayed in the right-hand area.
• From the MCP group, select the desired templates in the middle area of the page and click the green button Import templates on the right.
• Through the import, the templates become fully-fledged query reports or import interfaces and then appear directly in the respective areas — ready to use immediately.

Note: The templates contain the standard master data fields. You can add further master data fields manually or with the help of your AI assistant.

Our recommendation: Start with the ready-made templates instead of starting from scratch. Pull the appropriate template into your LDB, assign access rights to it, and activate it.

Use cases

Automatically checking ISO standards for currency

The AI combines the documents stored in your LDB and released via a query report with a free internet search, and checks which of your standards are up to date.

• Open your AI with an active connection to the LDB.
• First ask which standards exist (pure read access via the query report).
• Then ask which of your standards have been updated. The AI compares your controlled documents with the currently valid version and reports the deviations in a targeted way.

This way you receive only the information relevant to you and do not have to go through an extensive catalog of standards manually.

Figure: Document control

Automating the check as a scheduled task

So that recurring checks do not have to be triggered manually each time, they can be stored as a scheduled task. This way you keep the overview without having to start the query yourself every time.

• Instruct the AI to run the check regularly (e.g. "every Monday at 7 a.m.").
• Have a Scheduled Task created. The AI assistant then runs the query automatically and provides you with an overview of the points to be processed.

Filling in the pathology/biopsy form by voice

At the microscope — often while wearing gloves — typing on the computer is impractical. Using an AI skill and voice mode, the LDB pathology/biopsy form can be filled in hands-free.

• First instruct the AI step by step in the chat how to solve the task, until the process works.
• Have a Skill created from this. The skill is stored as a document in the background and describes what should be done. You can revise it if needed.
• Activate the skill, enter the sample number for which you want to fill in the pathology form, and activate voice mode.
• Answer the read-out options by voice. The AI structures the information and enters it via the import interface into the corresponding field of the sample.

The recorded information is then available for the report — e.g. as a design element or in the report table.

Figure: Pathology/biopsy form

Related topics

• Access rights
• Audit Trail: traceable documentation of all work
• Enable/disable AI integration