@androidosswal/mcp-sap-b1

An MCP (Model Context Protocol) server that gives AI coding assistants ground-truth knowledge of a SAP B1 Service Layer — schema (including custom UDFs) and a single latest sample row per entity — so generated code matches the customer's actual setup instead of hallucinating.

Read-only. Lean output. One process per project.

What it exposes

Exactly two tools:

list_endpoints

{ companyDB: string }
→ { count, endpoints: string[] }

All entity-set names from the Service Layer $metadata, sorted.

describe_endpoint

{ companyDB: string, entity: string }
→ {
    schema: { entitySet, keys, fields: [{ name, type, nullable?, isUDF?, maxLength? }] },
    latestRow: <one record, latest by CreateDate desc> | null
  }

The schema is parsed from $metadata and trimmed (no SAP boilerplate). The sample row is fetched live with $top=1&$orderby=CreateDate desc (fallback to first key). Binary fields are replaced with "<binary stripped>".

Configuration

Env vars (set in your MCP client config):

| Var | Required | Notes | |---|---|---| | SAP_URL | ✅ | Full base URL incl. /b1s/v1, e.g. https://sap.acme.com:50000/b1s/v1 | | SAP_USERNAME | ✅ | Service Layer login user | | SAP_PASSWORD | ✅ | Service Layer password | | SAP_CA_BUNDLE | ❌ | Path to a PEM CA bundle (for self-signed corporate CAs) | | SAP_INSECURE_SSL | ❌ | true to disable TLS verification entirely. Off by default. |

TLS is strict by default. Both insecure modes log a warning at startup when active.

Install / Run

Via npx (recommended)

In your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "sap-b1-acme": {
      "command": "npx",
      "args": ["-y", "@androidosswal/mcp-sap-b1"],
      "env": {
        "SAP_URL": "https://sap.acme.com:50000/b1s/v1",
        "SAP_USERNAME": "manager",
        "SAP_PASSWORD": "yourPassword",
        "SAP_INSECURE_SSL": "true"
      }
    }
  }
}

Local development

npm install
npm run dev   # runs src/index.ts via tsx

Build

npm run build   # emits dist/
npm start       # runs dist/index.js

Typical AI workflow

1. AI calls list_endpoints({ companyDB: "SBODEMOUS" }) once — caches the list mentally.
2. Per generated piece of code, AI calls describe_endpoint({ companyDB, entity: "Orders" }) to confirm the field set (especially U_* UDFs) and see one realistic row.
3. AI generates code that maps onto the real schema, not a generic SAP B1 template.

License

MIT