Backtest SPX dealer-gamma rules from your AI assistant
gex.live has an MCP server. Add it to Claude, Cursor, ChatGPT or any MCP client and the assistant can read the dealer-positioning archive and drive the backtesting Lab on your behalf. The one-line version is the title. Here is the rest. Two tiers, one rule: free data stays free Free, no key — the same finished-session data that is free on the website: list_sessions — finished SPX sessions in the…
The article outlines the capabilities of a proprietary AI assistant, hosted by gex.live, designed to analyze and backtest SPX dealer-gamma rules. The assistant requires a Lab token to access its tools, which are metered credits similar to those used on the site. It offers two primary functions: retrieving finished session data for analysis and compiling and running custom trading rules through a metered lab process.
Key features include:
- Data retrieval: Users can access finished SPX sessions with details such as OHLC, gamma flips, walls, hold bands, net-gamma percentile, and ATM IV. This data is free to retrieve but must be called via the `get_session` function, with a maximum of 50 sessions per call.
- Rule compilation: The `lab_compile` function allows users to transform verbal trading ideas into testable rules. The assistant returns a compiled specification, may ask clarifying questions, or provide compilation errors.
- Rule execution: The `lab_run` function executes the compiled rule through a backtesting process, followed by a quant optimization step using LightGBM. This process costs one credit and provides a truthful verdict on the rule's effectiveness, including performance metrics for different eras and a holdout period.
The assistant maintains a state of the user's Lab, including all active ideas, their stages, results, and the current credit balance, accessible via the `lab_state` function. Users can interact with their ideas through the `lab_idea` function, which allows actions such as updating, deleting, or adjusting the display/alert settings for specific ideas.
Unlike a traditional signal service or live feed, the Lab operates on data from completed sessions, not real-time market positioning. The assistant is designed to work with Claude, Cursor, ChatGPT, or any MCP client, with the option to add the MCP server as a transport for these clients. The service is accompanied by OAuth-based authentication for Cursor and the CLI, and a separate OAuth facade for web and desktop applications.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
