rudycanshoot: Instructions file for Codex

rudycanshoot — Agent Instructions

Project Overview

rudycanshoot is an MCP server and CLI tool that enables AI assistants to capture and view screenshots across Linux, macOS, and Windows.

MCP Server

Start with: node bin/rudycanshoot.js serve

Tools exposed:

• take_screenshot(mode, area?, filename?, outputDir?) → file path
• read_screenshot(path) → base64 image
• list_screenshots(limit?) → list of recent captures
• record_video(durationSec?, fps?, area?, temporary?, returnFrames?, maxFrames?) → video path + frames
• read_video(path, maxFrames?) → frames as images for AI viewing
• list_videos(limit?) / cleanup_videos(all?, olderThanMinutes?)

Coding Conventions

• ES modules (type: "module" in package.json), use import/export
• Node.js built-in APIs preferred over third-party where possible
• Platform detection via process.platform (linux, darwin, win32)
• Screenshot backends tried in order: prefer native/installed tools, fall back gracefully
• No TypeScript — plain JavaScript with JSDoc if types need documenting

Testing

node bin/rudycanshoot.js capture --mode fullscreen
node bin/rudycanshoot.js list

Adding a New AI Tool Installer

Edit src/install.js, add an entry to TOOL_INSTALLERS:

"my-tool": () => {
  const configPath = join(HOME, ".my-tool", "mcp.json");
  const config = readJson(configPath);
  config.mcpServers = config.mcpServers || {};
  config.mcpServers["rudycanshoot"] = MCP_ENTRY;
  writeJson(configPath, config);
},