A clean, single-agent travel planning system built with Claude API and direct tool integration (no MCP needed!).

Flub Agent is a simple, powerful travel planning assistant that uses Claude's native function calling with direct Python tool integration for flight search and other travel services.

• Single Agent Architecture: One Claude agent with direct access to all tools
• No MCP Complexity: Tools are simple Python functions - no servers to run
• Natural Tool Selection: Claude automatically uses the right tools for each query
• Conversation Context: Maintains history for multi-turn planning
• Easy to Extend: Add new tools by creating Python functions

• Python 3.10+
• An Anthropic API key
• (Optional) Bun or Node.js for iMessage integration

1. Clone and navigate to the repository

cd flub-agent

2. Install dependencies

# Python dependencies
pip install -r requirements.txt

# Optional: For iMessage integration
bun install

3. Set up environment variables Create a .env file:

ANTHROPIC_API_KEY=your_anthropic_api_key_here

4. Run the agent

# Test the agent
python main.py

# Or start the API server for iMessage integration
python agent_server.py

from src.simple_agent import SimpleFlubAgent

# Create agent
agent = SimpleFlubAgent()

# Ask a question
response = agent.process("What's the cheapest flight from EWR to LAX on 2025-11-10?")
print(response)

agent = SimpleFlubAgent()

# First query
response1 = agent.process("Find flights from NYC to LA on December 1st")
print(response1)

# Follow-up with context
response2 = agent.process("What's the cheapest option?")
print(response2)

# Clear history when done
agent.clear_history()

flub-agent/
├── src/
│   ├── simple_agent.py        # Main agent using Claude function calling
│   └── tools/                 # Tool functions
│       ├── __init__.py
│       └── flight_search.py   # Flight search tools
├── main.py                    # Test/demo entry point
├── agent_server.py            # HTTP API server for iMessage integration
├── imessage-watcher.ts        # TypeScript iMessage watcher
├── requirements.txt           # Python dependencies
├── package.json               # Node dependencies
├── .env                       # Environment variables (create this)
├── README.md                  # This file
├── CONVERTING_MCPS.md         # Guide for converting MCP servers
└── IMESSAGE_SETUP.md          # iMessage integration guide

User Query
    ↓
SimpleFlubAgent (Claude with Function Calling)
    │
    ├─ Tool: search_flights()
    ├─ Tool: find_best_price()
    └─ Tool: [Add more tools...]
    │
    ↓
Claude automatically selects and calls appropriate Python functions
    ↓
Unified Response

search_flights

• Search for flights between airports on a specific date
• Parameters: date, from_airport, to_airport, adults, max_results
• Returns: List of flights with pricing, duration, and details

find_best_price

• Find the cheapest flight for a route and date
• Parameters: date, from_airport, to_airport, adults
• Returns: Cheapest flight with price comparisons

To add a new tool:

1. Create a tool file in src/tools/:

# src/tools/weather.py
def get_weather(city: str, date: str):
    """Get weather forecast for a city on a specific date."""
    # Your implementation
    return {"temperature": 72, "conditions": "sunny"}

2. Export it in src/tools/__init__.py:

from .flight_search import search_flights, find_best_price
from .weather import get_weather

__all__ = ['search_flights', 'find_best_price', 'get_weather']

3. Register it in src/simple_agent.py:

# Import the tool
from tools import get_weather

# Add to self.tools list
self.tools.append({
    "name": "get_weather",
    "description": "Get weather forecast for a city on a specific date",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "City name"},
            "date": {"type": "string", "description": "Date in YYYY-MM-DD format"}
        },
        "required": ["city", "date"]
    }
})

# Add to _call_tool method
def _call_tool(self, tool_name: str, tool_input: Dict[str, Any]):
    if tool_name == "get_weather":
        return get_weather(**tool_input)
    # ... existing tools

That's it! No servers to run, no HTTP endpoints to configure.

SimpleFlubAgent(api_key: Optional[str] = None)

Methods:

• process(message: str) -> str: Process a message and return response
• clear_history(): Clear conversation history

Want to respond to text messages automatically? Check out IMESSAGE_SETUP.md!

Quick start:

# Terminal 1: Start the API server
python agent_server.py

# Terminal 2: Start the iMessage watcher
bun run watch

Now text your Mac: "What's the best flight from EWR to LAX tomorrow?"

See main.py for a working example:

from datetime import datetime, timedelta
from src.simple_agent import SimpleFlubAgent

agent = SimpleFlubAgent()

tomorrow = (datetime.now() + timedelta(days=1)).strftime("%Y-%m-%d")
response = agent.process(f"What is the best flight from EWR to LAX on {tomorrow}?")
print(response)

Required:

• ANTHROPIC_API_KEY: Your Anthropic API key

Make sure ANTHROPIC_API_KEY is set in your .env file.

Make sure you're running from the project root:

cd /path/to/flub-agent
python main.py

If a tool fails, check the error message in the response. The agent will report tool errors gracefully.

• Claude API Documentation
• Function Calling Guide

MIT License

For issues and questions, please open an issue on the repository.