AI Agent Crypto MCP Trading - Can AI make bag on Hyperliquid?

Detailed Summary

Introduction to AI Trading and Hyperliquid (0:00 - 1:08)

The video begins by showcasing a successful Solana short position, up 12.7% with 20x leverage, and then closes a Bitcoin trade for a 2.9% profit. It introduces the concept of using Large Language Models (LLMs) for crypto trading, referencing an experiment where an LLM named "Quen" achieved a 51% profit. The presenter then introduces Hyperliquid as the chosen platform for their AI agent, noting its API for programmatic trading.

Setting Up the Hyperliquid MCP Server (1:08 - 3:52)

• The presenter explains the creation of a Hyperliquid MCP server, which runs locally.
• Key tools integrated into the MCP server include get account info, get market data, open positions, place an order, close a position, and set leverage.
• The agent successfully retrieves the account balance ($93.97) and lists open positions using these tools.
• The underlying place order function, which executes requests to the Hyperliquid server, is briefly shown.
• The MCP server was built using 4,000 lines of documentation from the Hyperliquid platform.

Manual Trading Demonstration (3:52 - 8:44)

• The presenter demonstrates a manual trading workflow by prompting the AI agent.
• The agent first checks the account balance upon request.
• It then retrieves the latest pricing for Bitcoin, Ethereum, Solana, and BNB using parallel tool calling.
• The agent analyzes funding opportunities, suggesting BNB as the most interesting coin due to its -747 APR funding for shorts.
• After scanning all tradable coins, a 40x leverage Bitcoin long trade is suggested, which the presenter adjusts to 20x leverage.
• The agent also provides an explanation of perpetual futures trading when asked, differentiating it from CFDs.
• A 20x Bitcoin long trade is successfully placed and monitored, showing a quick profit of 0.2% initially, then $2 (2.9%) before being closed for profit.
• The account balance is re-checked, showing an increase to $94.

Automated Trading with '/cryptobro' (8:44 - 10:08)

• The video introduces an automated trading setup using a custom command /cryptobro.
• This command initiates an autonomous workflow that checks the current balance and open positions (noting a maximum of two open positions).
• The automated system identifies a new trading opportunity, suggesting a BNB long trade.
• An $80 margin, 10x leverage BNB long trade is placed autonomously.
• The Hyperliquid dashboard confirms the new BNB long trade is active.

Conclusion and Future Potential (10:08 - 11:03)

• The presenter highlights the educational value of the AI agent, allowing users to ask questions and learn about trading concepts while experimenting with small positions.
• The project is described as fun, responsive, and quick, offering opportunities for further development by adding more context and 'edge' to the trading strategies.
• The video concludes by recommending others to explore similar projects for learning about MCPs and crypto trading.

Notable Quotes

"So if you open up this, you can see we are 12.7% up on the Solana short 20x leverage uh position we we made here." — Presenter

"So lately there's been a lot of talk about using these LLMs to make uh trading in crypto." — Presenter

"I think this is pretty cool and we can do some Yeah, terminal trading here." — Presenter

"So what I think is pretty cool if you have some questions like you're pretty new to this, you can always ask, right?" — Presenter

"So you kind of learn about NCPs while you have some fun trading some crypto if you're into that and you just learn about stuff and ask questions on the run here." — Presenter

"So yeah, that was like a full cycle." — Presenter

"So what I think is good about this is it's a way nice way to learn about this because you can always ask questions on the way." — Presenter

"So it's pretty fun to play around with. And we can close this whenever you want." — Presenter

"So this is the place order function. This is the function that is executing the request from the hyperlquid server here on the tool place order." — Presenter

"So you can see as context here uh when I built this I collected 4,000 lines of uh documentation for the hyperlquid platform." — Presenter

"Quen makes 51% profit compared to other models." — Presenter

"So this is the same platform that they used in the experiment I think and they have an API." — Presenter