📈 Stock Analysis Agent with LangGraph

This repository documents my journey of learning LangGraph and building a production-ready Stock Analysis Agent that fetches real-time financial data, performs analysis, and integrates with Claude Desktop via the Model Context Protocol (MCP).
The work starts with exploratory notebooks and culminates in a deployable MCP server under stock-agent-mcp/.

---

📂 Project Structure

Stock_Analysis_Agent_LangGraph/
├─ .env.example                # Example env vars for root
├─ README.md                   # This README
│
├─ docs/                       # Screenshots & traces from LangSmith runs
│
├─ 1_simple_graph.ipynb        # Basics of LangGraph state graphs
├─ 2_graph_with_condition.ipynb # Adding conditional flows
├─ 3_chatbot.ipynb             # Simple chatbot agent
├─ 4_tool_call.ipynb           # Demonstrates calling external tools
├─ 5_tool_call_agent.ipynb     # Agent with tool orchestration
├─ 6_memory.ipynb              # LangGraph agent with memory
├─ 7_langsmith.ipynb           # Integration with LangSmith tracing
├─ 8_HITL.py                   # Human-in-the-loop example in Python
├─ 9_agent_real_time_stock_price.ipynb # Final stock analysis agent with tools
│
└─ stock-agent-mcp/            # Production-ready MCP server
   ├─ .env.example             # Example env for server
   ├─ README.md                # Server-specific README
   ├─ server.py                # MCP server entrypoint
   ├─ docs/                    # Demo media for MCP server
   └─ src/
      └─ langgraph_agent/
         └─ tools.py           # Tool definitions (stock price, earnings, etc.)

---

✨ Key Features

• LangGraph tool orchestration
• Alpha Vantage API for financial data (prices, earnings, company overview, historical)
• Gemini LLM for reasoning/orchestration (langchain-google-genai)
• LangSmith tracing (add env vars and it works automatically)
• HITL example for sensitive operations
• MCP server integration for Claude Desktop

---

🧭 Architecture (Notebook 9 – Final Agent)

flowchart LR
    U["User Query"] --> LLM["Gemini 2.0 Flash<br>(LangGraph Orchestrator)"]
    LLM --> T["Tools<br/>(search_ticker, overview, earnings, realtime, history)"]
    T --> AV["Alpha Vantage API"]
    AV -- JSON --> T --> LLM --> A["Final Answer"]

    %% Observability
    LS["LangSmith Tracing (optional)"]
    LLM -.-> LS
    T -.-> LS

Loading

---

🛠️ Tech Stack

• LangGraph • LangChain
• Gemini API (via langchain-google-genai)
• Alpha Vantage (financial data)
• LangSmith (optional tracing)
• Model Context Protocol (MCP)
• Env & packaging with uv

---

📸 Snapshot of Trace

Trace Explanation: This snapshot shows how the agent processes a query step by step — from the LLM’s reasoning, to tool calls, and finally producing the answer.

For the server demo video and GIF, see: stock-agent-mcp/README.md

---

⚡ Setup

Prereqs: Python 3.12+, uv installed

1. Clone & enter the repo

git clone https://github.com/shyamc757/Stock_Analysis_Agent_LangGraph.git
cd Stock_Analysis_Agent_LangGraph

2. Install dependencies (root workspace)

uv sync

3. Configure environment variables
   Root and server each provide a separate example file:

• Root: copy .env.example → .env (used by notebooks)
• Server: copy stock-agent-mcp/.env.example → stock-agent-mcp/.env

Populate the keys you have:

# Common (used across notebooks / server)
ALPHAVANTAGE_API_KEY=...
ALPHA_VANTAGE_BASE_URL=https://www.alphavantage.co/query

GEMINI_API_KEY=...

# Optional (enables tracing automatically if set)
LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT=https://api.smith.langchain.com
LANGSMITH_API_KEY=...
LANGSMITH_PROJECT=<project-name>

4. Run notebooks (root)

uv run jupyter lab

---

📬 Where to Look Next

• Server details, how to run, and demo media:
  stock-agent-mcp/README.md

---

📝 License

This project is licensed under the terms described in LICENSE.