What This Example Shows
- A focused single-agent build for a specialized domain (crypto quant analysis)
- How to write a long, structured
system_prompt that constrains output to quantitative reasoning
- How to wire the agent up to live market data via MCP (optional)
This is the same pattern as the Single Agent Overview, specialized for crypto. The same shape extends to any analyst domain — credit risk, equity research, intelligence triage, claims review. Step 1: Setup
import json
import os
import requests
from dotenv import load_dotenv
load_dotenv()
API_KEY = os.getenv("SWARMS_API_KEY")
BASE_URL = "https://api.swarms.world"
headers = {"x-api-key": API_KEY, "Content-Type": "application/json"}
Step 2: Define the Agent
The system_prompt is where domain expertise lives. Be specific about responsibilities and the kind of reasoning you want.
def run_crypto_analyst(task: str) -> dict:
payload = {
"agent_config": {
"agent_name": "Crypto Quant Analyst",
"description": (
"Quantitative analyst specializing in cryptocurrency markets "
"and trading strategies."
),
"system_prompt": (
"You are a Crypto Quant Analyst with deep expertise in "
"quantitative analysis of cryptocurrency markets. Analyze crypto "
"market data, identify trading patterns, calculate risk metrics, "
"and develop data-driven trading strategies.\n\n"
"Key responsibilities:\n"
"- Perform technical analysis on cryptocurrency price movements\n"
"- Calculate volatility, correlation, and risk metrics\n"
"- Identify market trends and trading opportunities\n"
"- Analyze trading volumes and market liquidity\n"
"- Provide quantitative insights for portfolio management\n"
"- Suggest risk-adjusted trading strategies\n\n"
"Always support your analysis with quantitative data and "
"statistical reasoning. Consider market cap, trading volume, "
"price volatility, asset correlation, and technical indicators."
),
"model_name": "gpt-4.1",
"role": "worker",
"max_loops": 1,
"max_tokens": 8192,
"temperature": 0.3,
},
"task": task,
}
response = requests.post(
f"{BASE_URL}/v1/agent/completions",
headers=headers,
json=payload,
timeout=120,
)
response.raise_for_status()
return response.json()
Step 3: Run It
if __name__ == "__main__":
result = run_crypto_analyst("Analyze Bitcoin (BTC).")
print(json.dumps(result, indent=2))
Adding Live Market Data via MCP
To give the agent access to real-time market data, attach an MCP server URL. Any OKX/Binance/CoinGecko MCP server works as long as it exposes the right tools.
"agent_config": {
# ...existing fields...
"mcp_url": "http://your-okx-mcp:8001/sse",
}
The temperature: 0.3 setting is intentional. Analytical and quantitative agents benefit from low-temperature outputs — they should be consistent and deterministic. Creative agents (copywriters, ideators) want higher temperatures. Pick the temperature to match the kind of reasoning you need, not the difficulty of the task.
Build Your Own Domain Agent
Replace the system_prompt and task to retarget:
| Domain | system_prompt focus | Suggested temperature |
|---|
| Credit risk | Default probability, debt service ratios, covenant analysis | 0.2 |
| Equity research | DCF, multiples, peer comp, catalysts | 0.3 |
| Threat intel triage | IOCs, attribution, severity scoring | 0.2 |
| Claims review | Coverage match, fraud signals, payout calc | 0.2 |
| Marketing copy | Tone, hook, CTA variants | 0.8 |
