Enable Autonomous Image Analysis for Agents

[tyrj8]

something like this works for now

MCP server that reviews your agent's UI edit using screenshots

This is an MCP server that reviews your agent’s edit by sending “before” and “after” screenshots to Qwen-72B. You can use any MCP server to take screenshots - I’ve been using browser-tools-mcp which has a takeScreenshot tool. Very useful to give claude a task and then have it make changes until the review passes. Basically TDD for UI? The cursor agent can’t “see” base64 images from MCP tool calls yet. Until Cursor adds image support for MCP tools, this is probably the most seamless way to ve…

can also just use claude instead of hyperbolic