Model Context Protocol (MCP)

Connect ChatGPT, Claude and other AI assistants to LDB

The Model Context Protocol (MCP) lets external AI applications access specific LDB functions that you have enabled. AI can then answer questions, retrieve current information, record structured data and carry out defined tasks.

Reports provide data from LDB. Import interfaces can send new data, tasks or status changes back to LDB.

  • A standard AI connection: MCP is an open standard supported by an increasing number of AI applications. This removes the need to develop a separate interface for each AI model.
  • Read and update: AI assistants can retrieve information through enabled reports and send data to LDB through defined import interfaces.
  • Targeted access: You decide which reports and import interfaces can be used through MCP. The AI does not receive unrestricted database access.
  • User-specific access rights: The MCP login is linked to an LDB user. The AI application can access only reports and import interfaces enabled for that user. All changes are recorded in the audit trail under the corresponding user and marked “MCP”.
  • Automated laboratory workflows: AI agents can support or automate recurring queries and clearly defined tasks.
  • Flexible model choice: The MCP connection works with different compatible AI applications and models.
Introducing MCP integration in the Fall Release 2026 user call

What is MCP?

The Model Context Protocol, or MCP, is an open standard for connecting AI applications to external data sources and functions.

In simple terms, MCP is a shared language between an AI and LDB. The AI can identify which enabled tools are available, what information they require and how results are structured.

In LDB, reports can be provided as MCP tools for reading data, and import interfaces as MCP tools for writing data. The AI therefore works through clearly defined interfaces rather than direct, unrestricted database access.

MCP is not an AI model. It connects AI applications such as ChatGPT or Claude to the LDB functions and data you have enabled for that purpose.

Example: voice-controlled sampling

During drinking water sampling, the sampling point, time, temperature, pH and other observations must be documented on site. Combined with an MCP-enabled AI voice application, this can be done hands-free.

For example, the sampler can dictate values through a suitably configured ChatGPT Voice Mode:

“Create a drinking water sample at the kitchen sampling point. Temperature 12.4 degrees, pH 7.3. No unusual odour or appearance.”

The AI structures the information and sends it to LDB through the enabled MCP import interface. LDB creates the sample and returns a status message with the sample number. The AI voice assistant reads out the status and number so that the sampler can label the bottle accordingly.

After sampling, the AI assistant can carry out further tasks:

  • Create a task for a deviation or a required follow-up check
  • Retrieve information about the next scheduled appointment
  • Show the next sampling point and the planned scope of testing
  • Identify missing information and ask targeted follow-up questions
  • Create a short summary of the sampling carried out

The sampler can keep their attention on the sampling point and their hands free for the sampling itself.

Control and security

MCP does not automatically give external AI access to all LDB data. Each report and import interface must be explicitly enabled for the user and for MCP use.

For sensitive actions or actions that write data, the AI agent should request confirmation before proceeding. This can be controlled through the associated MCP description.

See: General Data Security Guidelines (ADSR)
Step-by-step guide

MCP: Connect your LIMS to Claude and ChatGPT

Sebastian, Patrick and Konrad show you how to connect your AI assistant:

Onboarding October 2026

Connecting AI Assistants via MCP: Setup and Practical Examples

Patrick and Sebastian demonstrate setup, security, and three practical examples:

Questions about MCP integration and LIMS:

Yes. LDB connects to AI applications such as Claude, ChatGPT and Google Gemini through the Model Context Protocol (MCP). MCP is an open standard supported by all major AI providers, so you configure the connection once and can use it with any supported assistant. The AI works with current LIMS data: it retrieves enabled reports and returns new data through import interfaces. No separate programming is required for each AI application.
LDB LIMS supports MCP. With the Fall Release, the connection is available in production to all LDB customers from 7 September 2026, with user-specific permissions and a complete audit trail.
MCP (Model Context Protocol) is an open standard for connecting AI applications to external systems: a shared language between AI and LDB. Your laboratory no longer needs a separate interface for each AI tool. You enable individual reports and import interfaces, with control and traceability through the audit trail.
No. AI applications can access only functions and reports that have been specifically enabled. They do not receive unrestricted database access. LDB’s user-specific access rights also apply: the AI can see only what the signed-in user is allowed to see. Follow the principle of data minimisation by enabling only what the AI actually needs. Sensitive content can be excluded from enabled reports. LDB’s MFA and SSO also apply to MCP access. AI usage is disabled by default and must be deliberately enabled. Details on model selection and data protection are documented in our General Data Security Guidelines (ADSR).
Every MCP call is recorded in LDB against the relevant report and import interface, marked “MCP” and assigned to the user. Changes made through an AI application are therefore identifiable and attributable to a person at any time, with the same traceability as an entry made through the user interface. Your approval and validation steps remain unchanged. With MCP, the AI runs in your own AI application, such as Claude, ChatGPT, or Copilot. Whether data is used for training there is governed by your contract with the AI provider; business and enterprise plans typically exclude this. For the AI features built directly into LDB, on the other hand, using data to improve or train models is contractually excluded. This is documented in our General Data Security Guidelines (ADSR).
MCP is included at no additional cost in LDB licences with AI integration. Existing customers with an older licence without AI can take advantage of a discounted upgrade to the AI licence, which automatically enables MCP.
Open the report or import interface, click “Edit” under “Connection”, enable the MCP checkbox, enter the displayed MCP URL in your chosen AI assistant and authenticate once with your LDB credentials. The assistant can then immediately access the enabled data.
You will need the MCP server URL for your instance (https://labordatenbank.com/<instance>/mcp); it is displayed in the MCP section of every enabled report. Claude (claude.ai or Claude Desktop): Under “Customize” → “Connectors” → “Add custom connector”, enter a name and the server URL, then select “Connect” and allow access with your LDB login. For Team and Enterprise, an owner adds the connector for everyone. ChatGPT (chatgpt.com): Under “Plugins” → “+” → “Add custom MCP server”, enter a name and the server URL, select “OAuth” authentication, and sign in with your LDB login. In business workspaces, admins configure the server. Microsoft Copilot: In Copilot Studio, navigate to the agent under “Tools”, create a new tool of type “Model Context Protocol” with the server URL and “OAuth 2.0”, connect it with your LDB login, and publish the agent for Microsoft 365 Copilot or Teams; your IT team typically handles this. You only need to sign in once; after that, the enabled tools are ready in the chat. Menu names may vary slightly depending on the version. See the video guide to setting up LDB MCP in Claude
No. MCP is disabled by default and must be explicitly enabled for each report or import interface and each user. This conservative default protects against unintended data access.
No SQL knowledge is required to use MCP with an AI assistant: the AI handles the database queries. SQL knowledge is helpful only if you want to create your own reports or import interfaces to enable through MCP. Ready-made templates are available to get you started.
Ready-to-use MCP templates cover common use cases, including searches for sample orders, customers, prices, parameters and limit values, and imports of measurements, orders, customers and samples. Import them through System settings → AI integration → MCP templates → Import templates. The templates are then immediately available in the system.
MCP supports many laboratory automation use cases. Voice-controlled sampling: an AI voice assistant receives dictated sampling points, temperatures, pH values and observations on site, creates the sample through the import interface and reads back the sample number. Sample data maintenance: AI retrieves the sample, customer and order, adds missing measurements directly in the system and records everything in the audit trail. Capacity and staff planning: AI retrieves absence and leave data to help plan appointments or shifts with sufficient staff available.
No. MCP calls do not use AI tokens in LDB. Only the tokens of the respective AI assistant, such as Claude or ChatGPT, are used. An LDB licence with AI integration is required to use MCP.
LDB LIMS offers free workshops and onboarding sessions on AI in the laboratory, reporting and MCP.
Yes, that is precisely what MCP is designed for. The person responsible for your AI system enters the MCP server URL of your LDB instance (https://labordatenbank.com/<instance>/mcp) into your internal system. Depending on your organization, LDB may need to be approved centrally. Users sign in with their LDB credentials, and the permissions of each respective user apply.
When signing in, the AI assistant receives only a revocable key, never your password. Under Employees → API tokens, you can disconnect the integration at any time. For applications without a login dialog, personal API tokens can also be created there.
The AI decides which tool to use based on the tool name and description. A good description explains when to use the tool, what each parameter means, what it returns, and which tool makes sense next. Descriptive parameter names such as “Sample number” instead of “VAR0” also help. The AI can also draft a suitable description itself from the report’s SQL. Test new tools in a validation or demo instance first.

We use cookies

We use our own cookies and third-party cookies so that we can display the website to you and better understand how you use it, in order to improve the services we offer.