MCP server for SAP dumps

This will enable MCP clients to analyse SAP dumps. Inspired by mcp-windbg, explained in this blog

The idea is to enable AI tools to access dumps and ask details about them like description, call stack,...

The client usually needs a configuration file, like mcp.json for visual studio code, whose format depends on the application, see below for more details

The server can run as an independent process, connecting to clients over http(s) or as a subprocess

# start a server - you tipically want to do this independently, and give your clients the URL
npx -y sapdumpmcp
# start a server communicating over stdio - you tipically don't do it manually, you enter this in your client configuration
npx -y sapdumpmcp --stdio

demo

Claude desktop and similar

{
    "mcpServers": {
        "default-server": {
            "command": "npx",
            "args": [
                "npx",
                "-y",
                "sapdumpmcp",
                "--stdio"
            ],
            "env": {
                "ABAP_ADT_URL": "https://mysapserver.com",
                "ABAP_ADT_USER": "mysapuser",
                "ABAP_ADT_PASSWORD": "a secret password"
            }
        }
    }
}

Headers

When running as a server the SAP system details can be passed with headers:

• abap-server
• abap-user
• abap-password
• abap-language

The default values can be set as environment variables for debugging, shouldn't be used in production in webserver mode. A convenient way to inject them is with a .env file

Environment variables

Instead of using http headers, the server can be configured to use environment variables (perhaps stored in a .env file). It's the only option in STDIO mode Beware using these in http mode, as this will allow anyone with access to the server to run queries, so it's not recommended unless you restrict access in some other way The relevant variables are:

• ABAP_ADT_URL
• ABAP_ADT_USER
• ABAP_ADT_PASSWORD
• PORT

By default the server will listen to port 3000 when in http mode, nowhere in stdio mode

Testing

Running the develop launch configuration will:

• start the dev job - will serve the mcp server on port 3000 and restart it on code changes
• start the inspect job - will start an mcp inspector on port 6274, with a proxy on 6277
• start the debugger

you can test the server by connecting to the mcp inspector and select transport type Streamable HTTP and URL http://localhost:3000/mcp

Will only work if your .env file points to a valid server, as the inspector doesn't support headers based connections

Vscode/Github copilot

The mcp.json linked above works for visual studio code as long as:

• mcp is enabled (currently only available in preview)
• the dev job is running

To start the server you can use the command palette or the codelens on the mcp.json file: