Chromeflow

Guided web assistance for Claude Code and Codex CLI

Chromeflow — guided browser actions for Claude Code and Codex CLI Chromeflow connects your coding agent (Claude Code or OpenAI's Codex CLI) to your real Chrome browser. When the agent needs you to set up a third-party service, grab API keys, configure billing, or do anything in a web UI, Chromeflow drives it visually — in the browser you're already logged into. Why your real Chrome Tools like Playwright and Browser Use launch a fresh, empty browser — no cookies, no sessions, no 2FA. Chromeflow runs in your actual Chrome, where you're already signed into Stripe, AWS, Supabase, and everything else. The agent automates what it can, and pauses for the parts that genuinely need you. What it does - Highlights what to click with on-screen callouts - Clicks buttons, fills forms, uploads files, switches tabs, navigates pages - Reads page content as a human does — including React-controlled inputs, contenteditable editors (Stripe, Notion), CodeMirror, Monaco, same-origin iframes, and open shadow roots - Captures API keys and secrets and writes them straight to your .env - Pauses only for steps that genuinely need you — passwords, payment details, 2FA codes, personal choices How it works Chromeflow is two halves that work together: - This Chrome extension — acts on the active tab (highlights, clicks, fills, uploads) - A plugin for your agent (Claude Code or Codex CLI) — gives the agent an MCP server with the browser tools and a usage skill so it knows when to reach for them The agent drives the flow end-to-end. You watch and step in when something needs a human. Run Claude Code and Codex side-by-side in different Chrome windows — the popup shows which session is which with a host badge on each row. Things the agent can do with Chromeflow - "Set up Stripe — create a product with monthly and annual pricing, write the price IDs to .env" - "Open Supabase and grab my project's anon key and service role key" - "Configure SendGrid webhooks for this app and verify the test event lands" - "Fill out this 40-field onboarding form using the data in customer.json" - "Watch the Vercel deploy and tell me when it finishes — open the logs if it fails" Setup 1. Install the extension — click Add to Chrome above. 2. Install the plugin in your agent (one time, machine-wide). Claude Code: /plugin marketplace add NeoDrew/chromeflow /plugin install chromeflow Codex CLI: codex plugin marketplace add NeoDrew/chromeflow /plugins (then select chromeflow → install) 3. Restart the agent. That's it. The plugin registers the MCP server and ships the usage skill — no per-project setup. The agent will reach for Chromeflow automatically whenever a task needs browser interaction. What's new in 0.9.1 - Codex CLI support — same MCP server, same Chrome extension, install via `codex plugin marketplace add NeoDrew/chromeflow` - Host badge in the popup — each session card now shows whether it's a Claude Code or Codex session, so multi-agent workflows stay legible - All the 0.9.0 additions still apply: find_text / find_input / wait_for_text for cheap targeted discovery, react_set_input / react_call_prop for React-heavy apps, click_element until-clauses (until_selector, until_url_contains, until_text_contains) so clicks confirm themselves Requirements - Google Chrome - Claude Code (https://claude.com/claude-code) or Codex CLI (https://developers.openai.com/codex/cli) - Node.js 22 (the plugin's MCP server needs it)