pam-mcp-server

pam-mcp-server

An MCP server exposing the SAP Product Availability Matrix (PAM) data from de.itsfullofstars.sap.pam (webapp/model/PAM.json) as queryable tools, so an LLM client can look up product/version support status without needing the PAM app's UI.

It is read-only and makes no network calls — it only reads the local PAM.json file.

This is deliberately local-only, not a hosted service: real PAM data is SAP customer-licensed information that can never be bundled with this app or served publicly (see ADR 0006). Each user generates their own PAM.json locally (npm run pam:import, from their own SAP PAM CSV export) and runs this server against it on their own machine — there is no remote endpoint to connect to instead.

Tools

• pam_list_fields — lists every filterable PAM field (e.g. version, product, line, status, eomm, ...), its human-readable label, and whether it's matched exactly or by substring. Call this first if you're unsure which field name to use.
• pam_list_field_values — lists the distinct values a given field currently has, most common first (e.g. every status value that actually occurs, with counts). Useful to discover the exact spelling of a status/category/line value before filtering.
• pam_query_products — the main query tool. Accepts one optional filter per PAM field (a string or array of strings, OR'd within a field, AND'd across fields), plus limit/offset pagination and a response_format ("markdown" summary table or "json" full records).

How specific a filter is determines how many entries come back, by design:

Setup

npm install
npm run build

Running

npm start                 # runs dist/index.js over stdio
npm run dev                 # tsx watch mode, runs src/index.ts directly
npm run inspector            # opens the MCP Inspector against the built server

By default, the server reads ../webapp/model/PAM.json (relative to this package, i.e. the main app's bundled PAM data). Point it at a different file — e.g. a freshly re-imported or uploaded export — via the PAM_JSON_PATH environment variable:

PAM_JSON_PATH=/path/to/other/PAM.json npm start

See ../docs/architecture.md for the full PAM.json field mapping, and ../docs/index.md for how PAM.json itself is generated from an SAP PAM CSV export.

Using with Claude Code / another MCP client

Add it as a local stdio server, e.g. via the Claude Code CLI:

claude mcp add pam -- node /absolute/path/to/pam-mcp-server/dist/index.js

or in an MCP client's JSON config:

{
  "mcpServers": {
    "pam": {
      "command": "node",
      "args": ["/absolute/path/to/pam-mcp-server/dist/index.js"]
    }
  }
}

Project structure

pam-mcp-server/
├── package.json
├── tsconfig.json
├── src/
│   ├── index.ts        # entry point: creates the McpServer, registers tools, connects stdio transport
│   ├── data.ts          # loads/caches PAM.json
│   ├── fields.ts         # field metadata (names, labels, match mode) shared by all tools
│   ├── filtering.ts       # filter matching + distinct-value counting logic
│   ├── format.ts          # markdown table rendering + response truncation
│   ├── constants.ts        # CHARACTER_LIMIT, pagination defaults
│   └── tools/
│       ├── listFields.ts
│       ├── listFieldValues.ts
│       └── queryProducts.ts
└── dist/                 # built output (entry point: dist/index.js)