yunIO MCP Integration with Langdock

This article shows how to integrate the yunIO MCP server with Langdock to enable natural language access to live SAP data. With this setup, users can retrieve and interact with SAP data without direct SAP access, transaction code knowledge, or SAP GUI experience.

General Workflow

To integrate yunIO with Langdock via the MCP server:

This setup allows Langdock to securely access and execute yunIO services in SAP through MCP, enabling both interactive use cases and automated business processes.

Prerequisites

• Access to a Langdock account with administrator rights.
• yunIO services must have meaningful descriptions and be enabled for MCP, see Documentation: MCP Server - Setup in yunIO.
• The MCP endpoint of yunIO is exposed to the internet, see Documentation: MCP Server - Setup in yunIO.

MCP Server Integration in Langdock

To use yunIO services in Langdock, connect your yunIO MCP server as an integration. This makes the MCP tools (services) available to agents and workflows:

1. Open https://app.langdock.com/ and go to Integrations.
2. Click Add Integration > Start from scratch.
3. Select the integration type Connect remote MCP.
   
4. Enter the URL of your yunIO MCP server, e.g., sap-link.example.com:8177/mcp.
5. Select the authentication method None.
6. Click [Create and connect]. The configuration menu of the integration opens. This may take some time.

7. In the configuration menu, add the following custom headers:

   Header name	Header value	Description
   Authorization	Basic [base64 encoded username:password]	Authentication against yunIO. If yunIO uses access restrictions, enter Basic followed by your base64 encoded yunIO credentials in the format username:password.
   Content-Type	application/json	Specifies that the data you send in the request body is formatted as JSON.

8. Click [Save] to save the integration settings.

9. Click [Test connection] to check the MCP server connection. If successful, Langdock displays a list of MCP-enabled yunIO services.
10. Select the services you want to expose to Langdock and click [Save].

The MCP integration is now available for use in Langdock agents and workflows.

MCP Server Tools in Agents

Agents in Langdock are LLM-powered components that can interpret user input, make decisions, and call integrations or workflows. In the context of yunIO, agents act as an intelligent layer between user intent and MCP tools. They analyze requests and invoke the appropriate yunIO services via the MCP server.

To use the yunIO MCP server tools in a Langdock agent:

1. Open https://app.langdock.com/ and go to Agents.
2. Click [Create agent].
3. Enter a name and description of the agent.

4. Write instructions for the agent using:

   • Langdock Prompt Engineering Basics
   • Langdock Agent - Writing Instructions
   • Langdock Agent Configurator Template
   Example

   Persona:
   You are an SAP execution agent operating via a yunIO MCP server. You convert user intent into precise MCP tool calls. You are strict, accurate, and safety-focused. You only act through available MCP tools and never assume data. SAP is a critical transactional system; correctness and control are mandatory.
   
   Tasks:
   
   1. Interpret user intent and map it to the most appropriate MCP tool (yunIO service).
   2. For read operations: retrieve SAP data with correct filters and parameters. Clarify ambiguities before execution.
   3. For write operations: collect all required parameters, validate completeness, and request confirmation before execution.
   4. For modify operations: require clear object identification (key fields), validate changes, and confirm before execution.
   5. Never guess missing parameters. Always ask targeted follow-up questions.
   6. Prefer the most specific tool. Use tool metadata to guide selection.
   7. Execute only after inputs are complete and validated.
   
   Context:
   You interact with SAP exclusively via yunIO MCP tools. Each tool represents a predefined yunIO service, e.g., executing BAPIs or transactions in SAP, reading table data, etc. Only exposed MCP tools are usable. Tool parameters and descriptions define valid usage. Errors may occur due to missing permissions or invalid inputs. Treat all returned data as authoritative. Do not fabricate, infer, or modify data outside tool execution. All state-changing actions have real business impact and must be handled cautiously.
   
   SQL rules for parameter formatting:
   
   - Leading Zeros: SAP IDs (LIFNR, KUNNR) must always be 10 digits. Convert inputs like "1023" automatically to '0000001023'.
   - Date Format: SAP requires dates as 'YYYYMMDD'. Convert user input like "01.02.2026" or "1 February 26" automatically. Never send dots, dashes, or slashes.
   - No Escaping: Send strings without masking special characters.
   - Where Clause: 
       - Always use single quotes for values in the WHERE clause (e.g., NAME1 = 'Easy'). Never use double quotes (") or backslashes (\). 
       - For joined SAP tables, follow this format: MAKT~MAKTX LIKE 'Hard Drive%' AND MARC~WERKS LIKE '1000'.
       - For search logic, use the LIKE operator with wildcards: NAME1 LIKE 'SearchTerm%'. Always place % after the search term.
   
   
   Format:
   
   - For reads: return structured results (key fields, lists, or concise summaries).
   - For writes/updates:
       1. Summarize intended action (object, identifiers, values).
       2. Ask for explicit confirmation.
       3. After execution, return action, object, and status.
   - For missing data: ask specific questions.
   - For errors: return clear cause and next step.
   - Keep responses concise, structured, and unambiguous.
   

5. In the Actions section, click [Add action]. The "Select action" menu opens.

6. In the “Select action” menu, add integrations or workflows that expose yunIO MCP tools. These actions represent the MCP tools the agent is allowed to call.
   
7. Test the agent and refine its instructions and configuration as needed.

After successful testing, the agent can be shared with your organization.

Best Practices for Agents

• Limit available actions: Only expose the MCP tools required for the agent’s purpose to reduce risk and improve accuracy.
• Be explicit in instructions: Clearly define how the agent should behave for read vs. write operations.
• Enforce confirmations: Always require explicit user confirmation for state-changing actions in SAP.
• Use structured outputs: This improves reliability when agents are used inside workflows.
• Iterate with testing: Validate edge cases such as missing parameters, invalid inputs, and permission errors.

MCP Server Tools in Workflows

Workflows are used for complex automations and processes that require multiple steps, conditions, or actions. Depending on the configuration, they can be triggered manually or by defined events.

To use yunIO services in a workflow, use one of the following integration patterns:

• Use MCP tools via an agent
  To add an agent to your workflow, click [+ Add] > Agents and select an agent that uses the yunIO MCP server tools. The workflow calls the agent, and the agent invokes the appropriate MCP tools as needed. This approach enables:

  • Tool selection and orchestration via LLM reasoning
  • Dynamic decision-making (choosing which MCP tool to call at runtime)
  • Multi-step interactions with yunIO services

  For more information, see Langdock Documentation: Agent Nodes.

• Add MCP Tools as Action nodes
  To add a specific MCP tool to your workflow, click [+ Add] > Integrations and select the MCP tool. Configure the required input parameters for the tool. Use this approach when:

  • You want deterministic execution (no LLM decision-making)
  • The exact MCP tool and inputs are known in advance
  • You need predictable, repeatable workflow behavior

  For more information, see Langdock Documentation: Action Nodes.
  

• Alternative: Call services without MCP
  To call yunIO services directly without MCP, add an HTTPS request node to your workflow. Provide authentication, headers, and payload manually. This approach bypasses the MCP server and is useful if:

  • MCP is not available or not required
  • You need full control over the HTTP request

  For more information, see Integrate a yunIO Service with Langdock.

Best Practices for Workflows

• Error handling: Add conditional nodes or fallback paths to handle failed MCP calls.
• Logging & debugging: Use workflow logs to inspect MCP tool calls and responses during execution.
• Input mapping: When using action nodes, ensure inputs are correctly mapped from previous workflow steps.

Related Topics

• Langdock Documentation: MCP Integrations
• Langdock Documentation: Variable Usage
• Knowledge Base: Integrate a yunIO Service with Langdock
• Documentation: MCP Server
• Documentation: Server Settings