Python scripts exploring LangChain and LangGraph for AI agent development, with a focus on Pokémon card price analysis. These are standalone experiments and learning exercises — not production code.

LangGraph ReAct pattern — create_react_agent handles the think→act→observe loop automatically, but you need to be intentional about when to use tool-calling vs. direct completion.

InMemorySaver for conversation state — trivial to add multi-turn memory to a LangGraph agent, but the state is lost on process restart. For persistent memory, you'd need a database checkpointer.

FastAPI + LangGraph (agent-api.py) — wrapping an agent as an HTTP service is straightforward. The key insight: thread_id in the config is how LangGraph separates conversation state across concurrent users.

Direct Anthropic SDK vs. LangChain — price-analysis-agent.py intentionally bypasses LangChain and calls the Anthropic API directly. It's simpler for single-turn analysis tasks where you don't need the agent loop overhead.

pip install -r requirements.txt

Create a .env file:

ANTHROPIC_API_KEY=sk-ant-...
POKEMON_TCG_API_KEY=...     # Required for historic-pricing-script.py only

# Run basic agent
python langchain-agent.py

# Start agent API server
uvicorn agent-api:app --reload
# POST http://localhost:8000/chat
# {"message": "What's a Charizard worth?", "thread_id": "user-1"}

# Run price analysis (uses Anthropic SDK directly)
python price-analysis-agent.py

• Python 3.10+
• See requirements.txt

• agent-dev — Containerized dev environment (Docker + MCP servers) that these agent scripts can be run inside
• PCPC — The production version of the Pokémon price checker that informed the data analysis work in these experiments