Use

What the MCP Server Can Do

Use Case	Example
Fetch enterprise data	Pull sales order data from an SAP backend
Execute business processes	Create and run workflow jobs
Answer document questions	Hybrid QA using FAISS + LLM
Connect AI to live systems	Custom handlers for SAP, OData, and other backends

Architecture & Processing Flow

Client
  ↓
API Routers  (routers/upload_router.py)
  ↓
Core Manager
  ↓
Tool Loader → Tool Generator → Tool Executor
  ↓
Custom Handlers  (custom_handlers/)
  ↓
Backend Services  (AI Core, Vector Store, Chat History)

Component	File	What it Does
Tool Loader	core/tool_loader.py	Discovers and loads tool definitions and handler modules at startup
Tool Generator	core/tool_generator.py	Generates runtime adapters from tool manifests
Tool Executor	core/tool_executor.py	Validates inputs, runs the tool, enforces error handling and timeouts
YAML Validator	core/yaml_validator.py	Validates manifests and tool metadata
Auth Manager	core/auth_manager.py	Performs authentication and authorization for all API requests

API Usage

List Available Tools

Call this first to discover what capabilities are registered in the runtime.

curl --location 'https://cw-mcp-server.cfapps.eu10-004.hana.ondemand.com/mcp/{{env}}' \
--header 'Content-Type: application/json' \
--data '{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list"
}'

Execute a Tool

curl --location 'https://cw-mcp-server.cfapps.eu10-004.hana.ondemand.com/mcp/{{env}}' \
--header 'Content-Type: application/json' \
--data '{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "create_task_and_run_job",
    "arguments": {
      "user_id": "muthuveeramuni.v@incture.com",
      "process_name": "appcreation2",
      "name": "name8",
      "age": "Age8"
    }
  }
}'

Execution flow: Tool Loader → Validator → Executor → Handler → Backend Service → Response

REST Endpoints

Method	Endpoint	What it Does
POST	/upload	Accepts files and runs the document ingestion pipeline
GET	/status	Returns health and status information
POST	/mcp/{env}	JSON-RPC interface for tool discovery and execution

Authentication

Method	Description
API Key	Token passed via environment variable; validated on each request
Bearer Token	JWT or OAuth token passed in the Authorization header

Flow: client sends token → auth_manager validates → principal/claims attached to request context → authorization check runs before tool execution.

Hybrid QA Pipeline

Answers user questions by combining FAISS document retrieval with an LLM.

Step	What Happens
1. Retrieve	Query FAISS vector store for the top-N most relevant document chunks
2. Assemble	Build a context window from retrieved chunks using prompt templates
3. Generate	Send context + user query to the LLM via aicore_service.py
4. Post-process	Apply citations, provenance trimming, or re-ranking to the output

Key files:

File	Purpose
hybrid_qa/prompts.py	Prompt templates used for QA
hybrid_qa/service.py	Core QA orchestration logic
hybrid_qa/handler.py	Request handler for QA endpoints

Adding a New Custom Handler

1. Create a new module under custom_handlers/ (e.g., my_integration_handler.py)
2. Implement a handler class that inherits from custom_handlers/base.py
3. Register it by adding discovery metadata to manifest.yaml
4. Restart the server — the Tool Loader picks it up automatically

⚠️ Handler module names must exactly match what is declared in manifest.yaml. A mismatch will prevent discovery at startup.

Available Custom Handlers

File	Purpose
base.py	Base classes all handlers must inherit from
nwf_handlers.py	Native Workflow integration handlers
scp_handlers.py	SAP Cloud Platform handlers
soa_handler.py	Sales Order Automation handlers
odata_services_catalog.json	OData service metadata registry

Services Reference

Service	File	What it Does
AI Core	services/aicore_service.py	Orchestrates LLM calls and AI model logic
Vector Store	services/vectorstore_service.py	FAISS operations — persistence, indexing, retrieval
Chat History	services/chat_history_service.py	Persists and retrieves conversational context
Document Ingestion	services/document_ingestion.py	Preprocesses and ingests documents into the vector store