I built an MCP server for crypto trading strategies, design decisions and what I learned

Description and use

The MCP server streamlined crypto trading strategy development by separating signal logic from execution and leverage, and introduced a 7‑stage validation pipeline.

Problem

Developing algorithmic crypto trading bots was slow and error‑prone because each bot had to be manually backtested, tweaked, and redeployed. The loop of writing code, eyeballing results, and iterating 20+ times was brutal.

Outcome

The MCP server decouples strategy logic from execution and leverage, enabling rapid, automated development and rigorous validation. Bots can be deployed at any leverage without code changes, and the 7‑stage pipeline protects against overfitting.

Workflow

• 1. Define a strategy by subclassing Strategy and implementing on_bar(ctx).
• 2. Use the provided ctx to access OHLCV bars, feature catalog, and market regime.
• 3. Submit the strategy code to the MCP server via submit_strategy.
• 4. The server runs a 7‑stage validation pipeline (static check, in‑sample, out‑of‑sample, walk‑forward, randomized start, perturbation, holdout).
• 5. If the strategy passes all stages, it is deployed to the live platform.
• 6. The bot can be opened‑source, forked, or delisted via MCP tools.
• 7. Users can compete in tournaments and view performance via marketplace tools.

Evidence

The server exposes a strategy contract, a 7‑stage validation pipeline, and a set of tools for discovery, backtesting, submission, and marketplace interactions.

Source

Bots compete in 3‑day rounds across 6 domains (ETH/BTC/SOL spot + scalp).

Source