RAG-to-MCP

Nasscom AI-Code Sarathi · Developer Workshop R.I.C.E · CRAFT · sentence-transformers · ChromaDB · MCP

---

What You Will Build

Three use cases. One developer path. No branching.

UC	What you build	Core skill taught
UC-0A	Complaint Classifier	R.I.C.E prompt engineering + CRAFT loop
UC-RAG	Retrieval-Augmented Generation server	Chunking · Embedding · Retrieval · Grounded generation
UC-MCP	MCP tool server wrapping your RAG server	Tool contract design · JSON-RPC · Agent tool interface

Each UC builds on the previous. UC-MCP calls UC-RAG. UC-RAG uses the same policy documents that test UC-0A's edge cases.

---

The Stack

Component	Tool	Why
Embedder	sentence-transformers all-MiniLM-L6-v2	Local, no API key, ~80MB one-time download
Vector store	ChromaDB	pip install, no binary deps, works offline
LLM	Gemini free tier (swappable)	Free API key, no credit card, fast
MCP transport	Plain HTTP + JSON-RPC	No SDK — you see the raw protocol
Framework	R.I.C.E + CRAFT	Prompt discipline applied at every layer

---

Prerequisites

Install before the session — not during:

pip3 install sentence-transformers chromadb google-generativeai

Get a Gemini API key (free, no credit card):

1. Go to https://aistudio.google.com/app/apikey
2. Click "Create API key"
3. Set it: export GEMINI_API_KEY="your-key-here"

Verify setup:

python3 --version                    # Must be 3.9+
python3 -c "import chromadb; print('ChromaDB OK')"
python3 -c "from sentence_transformers import SentenceTransformer; print('ST OK')"

See docs/pre-session-install.md for the full setup checklist.

---

Repo Structure

RAG-to-MCP/
├── uc-0a/              Complaint Classifier — done with facilitator
│   ├── README.md
│   ├── agents.md       generate from README using AI
│   ├── skills.md       generate from README using AI
│   └── classifier.py   build using AI
│
├── uc-rag/             RAG Server — main developer UC
│   ├── README.md
│   ├── agents.md       generate from README using AI
│   ├── skills.md       generate from README using AI
│   ├── rag_server.py   build using AI
│   └── stub_rag.py     working fallback — used by UC-MCP
│
├── uc-mcp/             MCP Server — wraps your RAG server
│   ├── README.md
│   ├── agents.md       generate from README using AI
│   ├── skills.md       generate from README using AI
│   ├── mcp_server.py   build using AI
│   ├── test_client.py  pre-built — do not modify
│   └── llm_adapter.py  Gemini default — swap to Claude/OpenAI here
│
├── data/
│   ├── city-test-files/    test_pune.csv · test_hyderabad.csv · etc.
│   └── policy-documents/   policy_hr_leave.txt · policy_it_acceptable_use.txt
│                           policy_finance_reimbursement.txt
│
└── docs/
    ├── guide.md                 Step-by-step participant guide
    └── pre-session-install.md   Night-before install checklist

---

How to Submit

1. Fork this repo
2. Create branch: participant/[your-name]-[city]
3. Complete UC-0A, UC-RAG, UC-MCP
4. Commit after each UC using the formula:

   UC-RAG Fix [failure mode]: [why] → [what you changed]
   

5. Push and open a Pull Request against main

Your PR must contain agents.md, skills.md, and the .py file for all three UCs.

See docs/guide.md for the complete step-by-step guide.