MCP server

Alongside the UI5 app itself, the repository includes pam-mcp-server/ — a small MCP (Model Context Protocol) server that exposes the same PAM data (webapp/model/PAM.json) as tools an LLM client (Claude Code, Claude Desktop, or any other MCP client) can call directly, without going through the app's UI at all.

It's useful whenever you want an AI assistant to answer "what's the support status of X" questions against your own PAM data — e.g. while reviewing an architecture document, a dependency list, or a support ticket that mentions specific SAP product versions. See ADR 0007 for why this is a separate server rather than a feature of the UI5 app itself.

What it is

• Read-only. It only reads webapp/model/PAM.json from disk — no writes, no network calls.
• Local, stdio-based — deliberately not a hosted service. It runs as a local subprocess (node pam-mcp-server/dist/index.js) communicating over stdio. This isn't just a transport choice: real PAM data is SAP customer-licensed information (see ADR 0006 and CLAUDE.md → "Mock PAM data"), so it can never be bundled with this app or served from anywhere public — each user must generate their own PAM.json locally (npm run pam:import, from their own SAP PAM CSV export) and run the MCP server against it on their own machine. There is no remote/hosted endpoint to point an MCP client at instead.
• A thin, generic query layer over the same data the app uses. It does not duplicate PAM.json — both the UI5 app's pam model and the MCP server read the identical file (see Architecture → Data model & PAM.json), so they are always in sync with whatever npm run pam:import last produced.

See pam-mcp-server/README.md in the repository for the full project layout; the summary below focuses on how to use it.

Tools

Tool	Purpose
pam_list_fields	Lists every filterable PAM field (version, product, line, status, eomm, ...), its 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 field actually has in the loaded data, most common first (e.g. every real status string, with counts) — useful to discover an exact spelling before filtering.
pam_query_products	The main query tool: filters PAM entries by one or more fields (AND across fields, OR across multiple values for the same field), with pagination and a markdown/JSON output format.

How specific a filter is determines how many entries come back, by design — this directly implements the same "if it's the version, only that version; if it's the line, all entries of that line; if it's a status, every product with that status" behavior the app's own filters have:

You filter by...	You get...
version (e.g. "SAP S/4HANA 2025")	Typically a single matching entry.
product or line (e.g. "SAP S/4HANA")	Every entry belonging to that product/line.
status (e.g. "Unrestricted available")	Every product currently in that status.
Any other status/date field (ra_puna, ra_pua, ua, hanaStatus, hanaDate, eomm, eom, eop1, eosk, addOn)	Every entry with that exact value.

Setup

cd pam-mcp-server
npm install
npm run build

The repository's .mcp.json (at the repository root) already registers it as a project-level MCP server for Claude Code:

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

Since pam-mcp-server/dist/ is a build artifact (not checked into git), it must be built once (npm run build, above) before the server can start — Claude Code loads .mcp.json when a session starts, so a fresh checkout needs that build step before the tools show up (reload/restart the session afterwards if it was already running). See pam-mcp-server/README.md for configuring it in other MCP clients, and for pointing it at a different PAM.json via the PAM_JSON_PATH environment variable.

Worked example

This is an actual exchange from a Claude Code session with pam-mcp-server connected, included verbatim to show the full round trip: what the model asked the server, and what came back.

Prompt

lies die datei sad_sample.md. verwende den pam mcp sever um den supportstatus der darin erwähnten produkte herauszufinden. gib eine ganz kurze übersicht über den support status

(SAD_sample.md, at the repository root, is an unrelated sample Solution Architecture Document used as test fixture content elsewhere in this repo — it's not about PAM or this app at all. That's precisely what makes it a convenient real-world example here: it just happens to name concrete SAP product versions, the way a real architecture document, dependency list, or support ticket might.)

What the model did

Reading SAD_sample.md surfaces two concrete SAP product versions: the Target Platform (SAP S/4HANA 2022 On-Premise) and the backend technology stack's SAP NetWeaver 7.5 (hosting Adobe Document Services and the SAPUI5-based Fiori front end). The model called pam_query_products once per version, filtering by the exact version field:

pam_query_products({ version: "SAP S/4HANA 2022" })

{
  "total": 1,
  "count": 1,
  "offset": 0,
  "has_more": false,
  "entries": [
    {
      "version": "SAP S/4HANA 2022",
      "name": "SAP S/4HANA 2022",
      "product": "SAP S/4HANA",
      "line": "SAP S/4HANA",
      "category": "Enterprise Resource Planning",
      "addOn": "No",
      "status": "Unrestricted available",
      "ra_puna": "18.08.2022",
      "ra_pua": "12.10.2022",
      "ua": "12.10.2022",
      "hanaStatus": "Ready",
      "hanaDate": "",
      "eomm": "31.12.2027",
      "eom": "",
      "eop1": "",
      "eosk": "31.03.2030",
      "url": "https://userapps.support.sap.com/sap/support/pam?hash=pvnr%3D73554900100900005331"
    }
  ]
}

pam_query_products({ version: "SAP NETWEAVER 7.5" })

{
  "total": 1,
  "count": 1,
  "offset": 0,
  "has_more": false,
  "entries": [
    {
      "version": "SAP NETWEAVER 7.5",
      "name": "SAP NetWeaver 7.5",
      "product": "SAP NETWEAVER",
      "line": "SAP NetWeaver",
      "category": "Technology Platform",
      "addOn": "No",
      "status": "Unrestricted available",
      "ra_puna": "",
      "ra_pua": "12.10.2015",
      "ua": "20.10.2015",
      "hanaStatus": "Ready",
      "hanaDate": "",
      "eomm": "31.12.2027",
      "eom": "31.12.2030",
      "eop1": "",
      "eosk": "",
      "url": "https://userapps.support.sap.com/sap/support/pam?hash=pvnr%3D73554900100900000414"
    }
  ]
}

Final answer

Support-Status (via pam-mcp-server)

Die SAD referenziert SAP S/4HANA 2022 (Zielplattform) und SAP NetWeaver 7.5 (Basis von ADS/ NetWeaver Java, SAP_UI).

Produkt	Status	Ende Mainstream Maintenance	Ende Extended Maintenance
SAP S/4HANA 2022	✅ Unrestricted available	31.12.2027	– (Safekeeper-Service bis 31.03.2030)
SAP NetWeaver 7.5	✅ Unrestricted available	31.12.2027	31.12.2030

Beide aktuell unterstützt, Mainstream-Support endet in beiden Fällen Ende 2027.

What this demonstrates:

• The model used version (exact match) rather than product/line, since the document named specific product versions, not product families — exactly the "specificity drives result scope" design from Tools above.
• Both queries returned exactly one entry each (total: 1), letting the model read off status, eomm, and eom directly without any additional filtering or pagination.
• The raw JSON responses carry every PAM field (including ones the final answer didn't need, like url or hanaStatus); the model condensed that into just the fields the user's question actually called for — a markdown table of status and the two maintenance end dates.
• No PAM CSV re-import, no browser, no manual lookup on the SAP Support Portal was needed — the whole lookup happened through two tool calls against the locally generated PAM.json.