Build a Zero-Human Trading Firm with Paperclip + Claude Code
What you will build
A fully autonomous 5-agent AI trading firm using Paperclip and Claude Code — including a CEO agent, Research agent, Backtest agent, Risk agent, and Execution agent — managed through a Paperclip dashboard. **Prerequisites:**
Before you start
- Claude Code subscription (required)
- Paperclip account (free tier available)
- TradingView account (for market data)
- Basic understanding of trading concepts
- 15–20 minutes of uninterrupted setup time
Important Disclaimer
This guide is for educational and experimental purposes only.
Building AI systems that interact with financial markets carries significant real risk. Nothing in this guide constitutes financial advice, investment advice, or a recommendation to trade any financial instrument. Automated trading systems can and do lose money — including all invested capital.
Before using any automated system with real money:
- Consult a qualified financial advisor
- Understand the regulatory requirements in your jurisdiction
- Test extensively in paper trading (simulated) mode first
- Never risk capital you cannot afford to lose entirely
The Paperclip + Claude Code trading firm described in this guide is an experimental educational framework. Use it to learn about AI agent orchestration, not as a production trading system without extensive additional development and risk management.
What This Guide Covers
The Paperclip Zero Human Trading Firm, created by the jackson-video-resources GitHub project, is a one-shot prompt that builds a 5-agent AI organisation inside Paperclip using Claude Code.
What makes this interesting is not the trading — it is the architecture. It is one of the clearest demonstrations available of how AI agent orchestration works in practice: a CEO agent that manages a team, specialist agents with defined roles and responsibilities, structured reporting between agents, and autonomous decision-making within defined constraints.
Even if you have no interest in trading, the architecture here is a blueprint for how any multi-agent business operation could be structured.
The Five Agents
The trading firm creates five specialist agents:
CEO Agent — Your direct report. Manages the team, delegates research tasks, runs weekly board briefings, and reports to you on strategy and performance.
Research Agent — Scans market data sources, news, and analysis every night. Produces a weekly strategy brief with trade ideas for the Backtest Agent to evaluate.
Backtest Agent — Takes every trade idea from Research, runs historical backtests using TradingView data, logs every result, and passes validated strategies to the Risk Agent.
Risk Agent — Reviews every proposed trade against defined risk parameters. Maximum position size, maximum drawdown, correlation limits. Only passes trades that meet the criteria.
Execution Agent — Receives approved trades from Risk. In paper trading mode, logs the trade and tracks performance. In live mode (which requires additional broker API setup), executes the trade.
Step 1: Set Up Paperclip
Go to paperclip.ai and create an account. Paperclip is an AI agent management platform that handles the orchestration layer — how agents communicate, delegate tasks, and report to each other.
Connect your Claude Code subscription to Paperclip. In Paperclip settings, add your Anthropic API credentials.
Screenshot placeholder: Paperclip dashboard showing the connections settings
Step 2: Get the One-Shot Prompt
The prompt that builds the entire trading firm is available in the GitHub repository:
github.com/jackson-video-resources/paperclip-zero-human-trading-firm
Navigate to the prompts folder and open the main setup prompt. Copy the full text.
This is a long, detailed prompt that:
- Interviews you about your trading style and risk tolerance
- Installs the required components
- Creates the agent organisation structure
- Defines each agent’s role, responsibilities, and reporting relationships
- Configures the initial strategy parameters
Step 3: Run the Setup in Claude Code
Open Claude Code. Paste the full prompt.
Claude Code will begin by asking you a series of questions:
- Your trading style — swing trading, momentum, value, or mixed
- Asset classes — stocks, ETFs, crypto, or forex
- Risk tolerance — conservative, moderate, or aggressive
- Maximum position size — as a percentage of total portfolio
- Maximum number of concurrent positions
- Preferred trading timeframe — daily, weekly, or monthly holds
- Paper trading or live trading — strongly recommend paper trading to start
Answer each question honestly. The agents are configured based on your answers — wrong answers produce a system misaligned with your actual risk tolerance.
Screenshot placeholder: Claude Code terminal showing the interview questions
Step 4: Review the Agent Configuration
After the interview, Claude Code will generate:
- A Paperclip organisation structure with all five agents
- Role definitions for each agent
- Communication protocols between agents
- Initial strategy parameters based on your answers
- A paper trading tracking spreadsheet
Review the role definitions carefully. The constraints you set for the Risk Agent are the most important — this is your safety layer. Make sure the maximum position size and drawdown limits reflect what you actually want.
Screenshot placeholder: Paperclip showing the five-agent organisation structure
Step 5: Connect TradingView (Optional)
For the Research and Backtest agents to access market data, connect TradingView via the MCP (Model Context Protocol) integration.
In Claude Code, the TradingView MCP connection allows agents to:
- Pull historical price data for backtesting
- Access technical indicators
- Monitor watchlist alerts
To set up the TradingView MCP, follow the instructions in the GitHub repository README under the Integrations section.
Note: The system works in a limited capacity without TradingView — the Research Agent will use publicly available news and analysis instead of price data.
Step 6: Run Your First Board Briefing
Once the system is set up, ask the CEO Agent for a first board briefing. This is the first test of the system working as designed.
The CEO Agent will:
- Check in with the Research Agent for any pending analysis
- Check the Backtest Agent’s queue
- Check the Risk Agent’s current status
- Report back to you with a summary
If all five agents respond correctly, the system is working.
Screenshot placeholder: First board briefing output from the CEO Agent
What to Watch For
In the first week: Run the system in observation mode. Do not act on any trade suggestions. Watch how the agents communicate and whether the Research Agent’s output is coherent and useful.
In weeks two to four: If you are using paper trading, let the Execution Agent log its first trades and track how they perform against the Research Agent’s reasoning.
After one month: Review the Backtest Agent’s log. How many of the tested strategies passed the Risk Agent’s criteria? How are the paper trades performing?
Before ever going live: Consult a financial advisor. Understand your jurisdiction’s regulations around automated trading. Build in a human approval step for every trade if you proceed.
The Real Value of This Setup
The trading application is interesting. The architecture is more interesting.
What this system demonstrates is exactly how Taskmosphere builds multi-agent business operations: a management layer (CEO), specialist agents with defined scopes, structured reporting, and a risk/compliance layer that filters decisions before they execute.
The same architecture applies to sales teams, support operations, content production, and virtually any business function that involves multiple people making decisions within defined constraints.
That is the lesson here — not how to trade, but how autonomous agent systems work in practice.
Want someone to handle this setup for you?
We have done this setup hundreds of times. Book a free audit and we will scope exactly what your operation needs.