MCP SAP GUI Server

A Model Context Protocol (MCP) server for SAP GUI automation. This server provides tools to automate interactions with SAP GUI, enabling programmatic control of SAP transactions.

Requirements

- Python 3.8 or higher
- SAP GUI installed and configured
- Valid SAP credentials (system, client, user, password)
- Node.js (for npx)

Installation

0. Clone github folder and run the automated Install using setup script:

setup.bat

This will guide you through the build process and will integrate this directly in Cline or Roo, if you select so.

1. OR Manual Install using build script..:

build.bat

2. Configure SAP credentials:
- Copy .env.example to .env
- Update the values with your SAP credentials

3. Test server using mcp inspector:

run.bat debug

4. Use the integration script to automatically configure MCP settings:

integrate.bat cline  # Configure for Cline

integrate.bat roo    # Configure for Roo

The script will:
- Automatically determine the correct settings file path
- Create a backup before making any changes
- Safely update the MCP configuration
- Validate changes to prevent corruption

Manual Configuration (if needed):

    "mcp-sap-gui": {

      "command": "python",

      "args": [

        "-m",

        "sap_gui_server.server"

      ],

      "cwd": "PATH_TO_YOUR_FOLDER/mcp-sap-gui",

      "disabled": false,

      "autoApprove": []

    }

5. Use this prompt to explain the Tool to your AI Model:

Important Safety Notice:

SAP is a highly sensitive system where incorrect interactions can have serious consequences. Every action must be performed with utmost precision and care. When in doubt about any action, STOP immediately and request user assistance.

Available Tools:

The mcp-sap-gui server provides tools for SAP GUI interaction:

 launch_transaction: Start a new transaction

 sap_click: Click at specific coordinates

 sap_move_mouse: Move mouse to coordinates

 sap_type: Enter text into fields

 end_transaction: Close the current transaction

Technical Limitations and Requirements:

1. You will receive only screenshot images of the SAP GUI window after each action

2. No direct access to screen element metadata or technical representation

3. You must use image recognition to:

    Identify UI elements (fields, buttons, etc.)

    Determine precise x/y coordinates for interactions

    Verify element sizes and positions

4. All coordinates must be exact - approximate clicking is not acceptable
Step-by-Step Process:

1. Start SAP GUI Session:

    Call launch_transaction with desired transaction code

    Analyze the returned screenshot carefully

2. Interact with Screen:

    Use image recognition to identify needed elements

    Calculate exact coordinates for interaction

    Execute appropriate action (sap_click, sap_type, etc.)

    Verify result in next screenshot

3. Capture Screenshots:

    Save screenshots at key points in the process

4. End Session:

    Call end_transaction when finished
Best Practices:

1. Always verify screen state before any action

2. Double-check coordinates before clicking

3. Document each step with clear annotations

4. If uncertain about any element position, request user verification

5. Maintain consistent screenshot naming convention

Available Tools

The MCP SAP GUI Server provides the following tools for SAP automation:

Transaction Management

- launch_transaction: Launch a specific SAP transaction code - end_transaction: End the current SAP transaction

Interface Interaction

- sap_click: Click at specific coordinates in the SAP GUI window - sap_move_mouse: Move mouse cursor to specific coordinates - sap_type: Type text at the current cursor position - sap_scroll: Scroll the SAP GUI screen (up/down)

Screen Capture

- save_last_screenshot: Save the last captured screenshot of the SAP GUI window. Returns the absolute file path of the saved image.

Screenshot Return Formats

All tools that interact with the SAP GUI window (launch_transaction, sap_click, sap_move_mouse, sap_type, sap_scroll) support different screenshot return formats controlled by the return_screenshot parameter:

1. none (Default): Only returns success/error messages

{

    "type": "text",

    "text": "Status: success"

}

2. as_file: Saves screenshot to the specified target folder and returns the path

{

    "type": "text",

    "text": "Screenshot saved as C:/path/to/file/screenshot.png"

}

Note: When using as_file, you must specify the target folder using the as_file_target_folder parameter. The folder will be created if it doesn't exist.

3. as_base64: Returns the raw base64 string

{

    "type": "text",

    "text": "base64_encoded_string_here"

}

4. as_imagecontent: Returns MCP ImageContent object

{

    "type": "image",

    "data": "base64_encoded_string_here",

    "mimeType": "image/png"

}

5. as_imageurl: Returns embedded resource with data URL

{

    "type": "resource",

    "resource": {

        "uri": "application:image",

        "mimeType": "image/png",

        "text": "data:image/png;base64,..."

    }

}

Example usage:
```python