Building an Autonomous AI Trading Agent with Claude + TickAtlas

The Challenge

LLMs like Claude are powerful reasoning engines but they cannot access real-time market data. Feeding raw OHLCV data or indicator numbers to an LLM requires complex prompt engineering to make the data interpretable. The developer needed a way to give Claude "market vision" without building a custom data interpretation layer.

The Solution

The /v1/summary endpoint was the key. It returns natural language market analysis that aggregates all 42 indicators into a human-readable (and LLM-readable) summary. Claude can directly interpret this output without any data transformation or custom parsing logic.

# The AI agent's data source
response = requests.get(
    "https://tickatlas.com/v1/summary",
    headers={"X-API-Key": API_KEY},
    params={"symbol": "EURUSD", "timeframe": "H4"}
)

# Feed directly to Claude
analysis = response.json()["data"]["summary"]
claude_response = client.messages.create(
    model="claude-sonnet-4-20250514",
    messages=[{
        "role": "user",
        "content": f"Based on this analysis, should I trade? {analysis}"
    }]
)

The Results

• Fully autonomous analysis: The agent independently queries market data, interprets the summary, and generates trade recommendations every 4 hours.
• No manual parsing required: The /v1/summary output is already formatted for LLM consumption, eliminating the need for custom indicator interpretation.
• Low cost: At only 2-3 summary calls per day across 5 pairs, the Starter plan provides more than enough capacity.
• Reasoning transparency: Claude explains its reasoning step by step, citing specific indicator readings from the summary.

Architecture

Scheduler (every 4h) → Python agent
  → /v1/summary (EURUSD, GBPUSD, USDJPY, XAUUSD, BTCUSD)
  → Claude API (analyze + decide)
  → Slack notification (trade idea + reasoning)
  → Human reviews and executes

Key Endpoints Used