Use Chrome DevTools Protocol to Control WinForm/MFC Apps
Maps CDP to native Windows UI so agents skip browser limits.

118 CDP commands in a single Go binary for AI agents.
AI agent developers, automation engineers
Puppeteer · Playwright · cdp-cli
Each command prints JSON to stdout and uses semantic exit codes (0 success, 1 error, 2 can't connect, 3 timeout), so they compose naturally with pipes and &&.
I built it because I wanted AI agents to be able to do everything Chrome DevTools can do, not just the basics. Existing options were either too narrow (MCP browser tools cover click/navigate/fill but not network interception or heap snapshots) or too heavy (Puppeteer/Playwright are great but they're libraries — an agent has to write and run a script rather than just call a command).
There's no session to manage and no browser object to keep alive. An agent just runs hubcap goto --wait https://example.com && hubcap text '#price'.
It also includes a Claude Code skill. The skill only loads command categories into context and pulls full docs on demand via hubcap help, so an agent gets coverage of 118 commands without the context cost.
Source: https://github.com/tomyan/hubcap Docs: https://hubcap.tomyandell.dev
Maps CDP to native Windows UI so agents skip browser limits.
Discord bridge to local AI using CDP—elegant security model, but niche use case.
Agent-first browser automation via CDP, not a Playwright wrapper or subscription.
CDP debugger primitives enable function hooks without monkeypatching the runtime.
Exposes Chrome's locked Gemini Nano as an OpenAI-compatible server without API keys.
Slashes tokens vs Playwright by returning a DOM brief instead of full tree.