@qubittron/mcp-server-sap

MCP server for Qubi SAP — leverage Qubi's comprehensive SAP knowledge base as a knowledge retrieval layer for your AI coding assistant. Works with Claude Code, Claude.ai, Cursor, and any MCP-compatible tool.

How It Works

Qubi MCP provides SAP domain knowledge directly to your AI coding assistant:

1. You ask a SAP question or request
2. Your AI assistant (Claude Code, Cursor, etc.) calls Qubi MCP tools
3. Qubi searches its SAP knowledge base via vector similarity
4. Raw documentation is returned to your AI assistant
5. Your AI reasons over the SAP docs and produces the answer/code

Zero LLM cost on Qubi's side — all AI reasoning happens in your existing AI assistant. Qubi only provides the SAP knowledge retrieval.

Quick Start

Claude Code

claude mcp add qubi-sap -- npx @qubittron/mcp-server-sap

# Set environment variables
export QUBI_SERVER_URL=https://your-instance.qubittron.com
export QUBI_API_KEY=your-api-key

Claude.ai Desktop

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "qubi-sap": {
      "command": "npx",
      "args": ["@qubittron/mcp-server-sap"],
      "env": {
        "QUBI_SERVER_URL": "https://your-instance.qubittron.com",
        "QUBI_API_KEY": "your-api-key"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "qubi-sap": {
      "command": "npx",
      "args": ["@qubittron/mcp-server-sap"],
      "env": {
        "QUBI_SERVER_URL": "https://your-instance.qubittron.com",
        "QUBI_API_KEY": "your-api-key"
      }
    }
  }
}

Available Tools (8)

All tools return raw SAP documentation chunks. Your AI assistant synthesizes them.

| Tool | What It Searches | Best For | |---|---|---| | search_sap_knowledge | Full SAP KB (90+ modules) | General SAP questions | | ask_sap_agent | Module-filtered KB | Focused domain questions | | analyze_sap_project | Best practices for tech stack | Project review | | generate_cap_files | CAP patterns and examples | Code generation reference | | get_sap_best_practice | Recommended patterns | Pre-coding guidance | | validate_sap_code | Standards and error patterns | Post-coding validation | | get_sap_config_guide | Configuration documentation | System setup | | list_sap_agents | Static agent list | Discovery |

Tools Detail

search_sap_knowledge

Semantic vector search across the entire SAP knowledge base. Returns structured JSON with relevance scores and quality tier.

query: "How to configure batch jobs in S/4HANA"
module: "MM" (optional filter)
limit: 5 (1-20)

ask_sap_agent

Search SAP documentation with agent-specific module filtering. Automatically generates multiple search queries for comprehensive coverage.

question: "Best practices for CAP service handlers"
agent: "sap-cap" (optional, searches all modules if omitted)
context: "Using Node.js runtime with HANA" (optional)

analyze_sap_project

Get SAP best practice documentation relevant to a project's technology stack. Provide your project structure and key files for comparison.

projectStructure: "db/schema.cds\nsrv/service.cds\n..."
files: [{ path: "db/schema.cds", content: "..." }]
focusAreas: ["security", "deployment"]

generate_cap_files

Get SAP CAP documentation, patterns, and examples relevant to your requirements. Returns reference material for your AI assistant to use when generating project files.

requirements: "Timesheet management with employees and approval workflow"
entities: ["Employee", "Timesheet", "Approval"]
includeFiori: true
includeDeployment: true

get_sap_best_practice

Get SAP best practices and recommended patterns for a specific topic.

topic: "CAP service handler error handling"
module: "CAP" (optional)
area: "development" (optional)

validate_sap_code

Get SAP standards and common error patterns for validating code. Your AI assistant compares the provided code against the returned standards.

code: "using { managed } from '@sap/cds/common'; ..."
language: "cds"
fileType: "schema" (optional)
filePath: "db/schema.cds" (optional)

get_sap_config_guide

Get SAP configuration documentation for a specific setup task.

scenario: "Set up automatic payment program F110"
system: "S/4HANA" (optional)
module: "FI" (optional)

Claude Code Workflow Examples

1. Build a CAP app from scratch

> Use generate_cap_files to get patterns for a leave management system with Employee, LeaveRequest, and Approval entities, then write the files based on those patterns

2. Review an existing project

> Use analyze_sap_project with my project structure and key CDS files, focus on security and deployment readiness

3. Research before implementing

> Use search_sap_knowledge to find how draft-enabled services work in CAP, then get_sap_best_practice for CAP draft handling patterns

4. Get configuration help

> Use get_sap_config_guide for setting up OAuth2 SAML Bearer assertion in Integration Suite for S/4HANA connectivity

5. Validate code before committing

> Use validate_sap_code on my db/schema.cds and srv/service.js files before I commit

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | QUBI_SERVER_URL | Yes | Your Qubi instance URL | | QUBI_API_KEY | Yes | API key from Qubi dashboard |

Alternative variable names supported: QUBI_URL, QUBI_TOKEN

API Endpoints Used

The MCP server only calls two Qubi endpoints:

| Endpoint | Method | Purpose | |----------|--------|---------| | /api/sap/knowledge/search | POST | Vector similarity search | | /api/sap/health | GET | Health check |

No LLM endpoints are called. The /api/sap/chat endpoint is not used by the MCP server.

SAP Module Coverage

90+ SAP modules including: FI, CO, MM, SD, PP, QM, PM, PS, HCM, WM, TM, BTP, CAP, Fiori, HANA, Cloud ALM, Ariba, SAC, Datasphere, Integration Suite, SuccessFactors, RAP, ABAP, UI5, EWM, and more.

Development

cd packages/mcp-server
npm install
npm run build   # Compile TypeScript
npm run dev     # Start with tsx for development

Changelog

[2.0.0] - 2026-05-07

Changed (BREAKING)

• All tools now return raw SAP documentation chunks instead of LLM-generated responses
• Zero LLM cost on Qubi side — AI reasoning handled by your coding assistant
• Removed dependency on /api/sap/chat endpoint
• Only uses /api/sap/knowledge/search (vector search)

Added

• Multi-query search for comprehensive coverage
• Intelligent query generation based on question type (30+ SAP module keyword patterns)
• Agent-to-module mapping for focused searches
• Shared formatChunks() helper for consistent output

Removed

• askAgent() client method (replaced by searchMultiple())
• SSE response parsing (no longer needed)
• LLM JSON extraction utilities (no longer needed)
• LLM-generated file parsing (no longer needed)

[1.0.1] - 2026-05-04

• Initial release with 8 MCP tools
• LLM-powered responses via /api/sap/chat

Get Access

Contact [email protected] to get your Qubi platform URL and API key.