India AI Impact Summit 2026

Your Money.
Your AI.
Your Control.

Session 3 — Build an AI Finance Agent that connects to Zerodha via MCP. Live data. Local models. Zero exposure.

₹0
API fees
Live
Market data
0
Data shared
MCP Finance Agent
The Problem

Your portfolio.
Their servers.

  • Cloud AI sees your entire portfolio — holdings, positions, P&L, trading patterns.
  • Financial data is the most sensitive data. Sharing it with 3rd party AI is a compliance nightmare.
  • Every prompt to ChatGPT/Gemini is stored, logged, and potentially used for training.
❌ What happens today

You → ChatGPT → "Analyze my portfolio"

  • Your holdings sent to OpenAI servers
  • Your P&L data stored in their logs
  • Your trading strategy exposed
  • No control over data retention
The Solution

Run the AI
on your machine.

Use Intel AI Superbuilder with a local LLM + Zerodha MCP to get live market data without sending a single byte to the cloud.

  • 100% Local LLM — Your AI runs on your device's NPU/GPU.
  • Direct API connection — MCP talks to Zerodha, not a middleman.
  • Zero data leakage — Financial data never leaves your PC.
✅ What we're building

You → Local LLM → MCP → Zerodha API

  • AI runs entirely on your AI PC
  • MCP connects directly to Zerodha Kite
  • Live prices, positions, P&L — all local
  • Your data stays on your device
Key Concept

What is an AI Agent?

A chatbot can only think. An Agent can think AND act.

Chatbot (LLM)

Brain in a jar. Can answer questions but cannot do anything in the real world.

AI Agent

Brain + hands. Can reason, plan, and take actions via tools (APIs, databases, etc).

Our Finance Agent

LLM + Zerodha MCP. Reasons about finance, then fetches live data to answer you.

Key Concept

What is MCP?

Model Context Protocol — the universal adapter that lets AI talk to external services.

  • Open standard — Created by Anthropic, adopted industry-wide.
  • Plug & Play — Connect to APIs, databases, files. No custom code.
  • Secure — Data flows directly between your LLM and the API. No middleman.
// Think of MCP like USB-C
// One standard port. Many devices.

AI Superbuilder
  └─ MCP → Zerodha (stocks)
  └─ MCP → Gmail (emails)
  └─ MCP → Google Drive (files)
  └─ MCP → Slack (messages)
  └─ MCP → Any API...
💡 Analogy: MCP is like USB-C for AI. One universal connector, every tool.
Architecture

How It Works

You ask a question in English. The Agent figures out what API to call, fetches live data, and explains the answer.

You
"What's TCS at?"
Local LLM
Understands intent
MCP
Routes to API
Zerodha
Live market data
Answer
"TCS: ₹3,872 ↑1.2%"
Deep Dive

The Agent Loop
(ReAct Pattern)

The Agent doesn't just answer. It Reasons → Acts → Observes → Repeats until it has the full picture.

🔁 ReAct = Reasoning + Acting. The LLM decides which tool to use, calls it, reads the result, and decides what to do next. 
// User asks a complex question:
"Compare RELIANCE and TCS — which is
more volatile this week?"

Step 1 — Reason:
  "I need prices for both stocks"

Step 2 — Act:
  → call get_quote("RELIANCE")
  → call get_quote("TCS")

Step 3 — Observe:
  "RELIANCE: ₹2,845 (±3.2%)"
  "TCS: ₹3,872 (±1.1%)"

Step 4 — Answer:
"RELIANCE is more volatile at ±3.2%
vs TCS at ±1.1% this week."
Live Demo

Ask Your
Finance Agent

Type natural language queries. The Agent decides which Zerodha API to call, fetches live data, and explains.

// Try these prompts:

Level 1 — Single stock:
"What is TCS trading at right now?"

Level 2 — Comparison:
"Compare INFY, TCS, and WIPRO.
Which has the highest volume today?"

Level 3 — Analysis:
"Show me NIFTY 50 and Bank NIFTY.
Are we in a bullish or bearish trend?"

Level 4 — Portfolio:
"Show my portfolio positions and
calculate my total unrealized P&L"
Finance Agent Demo
🔍 Watch the Thoughts panel — you'll see the LLM reasoning step-by-step before calling each Zerodha API.
Live Scenario

The Morning
Market Brief

You're a fund manager. It's 9:30 AM. You need a quick overview of your portfolio + market sentiment before your first client call at 10 AM.

// Morning Brief Prompt

"Good morning. Give me:
1. Current NIFTY 50 and SENSEX levels
2. My top 5 holdings with today's change
3. Which sector is performing best today
4. Any stock in my portfolio that dropped
   more than 2% — flag it as a risk"

// The Agent will:
  → Fetch indices via MCP
  → Pull portfolio positions
  → Analyze sector performance
  → Flag risks automatically
NIFTY 50
22,456
↑ +0.8%
SENSEX
73,891
↑ +0.6%
RELIANCE
₹2,845
↑ +1.2%
⚠ HDFC BANK
₹1,621
↓ -2.4%
🚨 AI Alert: HDFC Bank dropped 2.4% — review before client call.
Session Complete

What You Learned

You built an AI Finance Agent that acts on real-world data — all locally.

AI Agents

LLMs that reason and act via tools

MCP Protocol

USB-C for AI — one standard, every tool

Live Market Data

Zerodha Kite quotes via natural language

Financial Privacy

Zero portfolio data leaves your device

1 / 10