
Superpowers Chrome
https://github.com/obra/superpowers-chrome- Category
- AI Agents
- Rank
- No. 643Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- obra
- GitHub
- 349 stars
- Latest release
- v3.0.5
- Date
About
Claude Code plugin that drives Chrome directly via the DevTools Protocol — zero dependencies, instant browser control.
What it does
Superpowers Chrome gives agents two ways to operate an existing Chrome session: a command-line interface with tab, navigation, interaction, extraction, screenshot, and raw protocol commands, or one MCP tool that dispatches browser actions. It can preserve profiles, run multiple profiles concurrently, and capture page state after DOM-changing actions.
Why it's ranked here
The appeal is focused control with little setup overhead. Tab indexes hide WebSocket endpoint details, while the MCP surface compresses many browser actions into one tool. Automatic HTML, Markdown, screenshot, console, and DOM-summary captures give agents useful state after changes. Dialog handling also prevents common prompts and device choosers from freezing the session.
What's good
It covers routine browsing without giving up lower-level access. Agents can navigate, wait, click, fill, select, extract content, take screenshots, and send raw protocol commands. Dynamic port allocation reduces local conflicts, named profiles support persistence or deliberate sharing, and separate default profiles let parallel MCP processes control independent Chrome instances. Page prompts become explicit synthetic responses with accept, dismiss, credential, and device controls.
Tradeoffs
Mode changes restart Chrome, reload tabs with GET requests, and lose POST state. They may also create extra blank tabs. A profile cannot change while Chrome is running. Headed Linux and WSL2 sessions require display configuration, while headless containers may need extra Chrome flags. Console capture is described as incomplete. The project itself recommends another automation stack for fresh browser instances, PDFs, and higher-level workflows.
How to use it well
Use it for agent workflows that need persistent browser state, direct tab control, compact MCP context, or inspectable captures after each interaction. Choose the command-line mode for explicit scripting and the MCP mode for clients that benefit from one browser tool. Assign fixed profile names when processes should share Chrome, and separate profiles when isolation matters. Do not treat it as the documented choice for fresh instances, PDF-heavy work, or higher-level automation abstractions.
Technical notes+
README.md defines the two interfaces, 17 command categories, profile and port behavior, dialog grammar, capture artifacts, and platform guidance. package.json declares an ESM package requiring Node 18 or newer, exposes a command-line binary, includes only development dependencies, and runs Biome, a bundle-freshness check, and Node tests. docs/cdp/flatten-mode.md describes one browser-level WebSocket with flattened session routing. docs/cdp/target-lifecycle.md documents target discovery, optional automatic attachment, and isolated browser contexts. docs/cdp/navigation-listener-race.md records listener-before-navigation ordering. docs/TESTING-INSTRUCTIONS.md notes POST-state loss during mode switching, possible blank-tab multiplication, required restarts for profile changes, and incomplete console capture.
Observed
- License
- MIT
- Primary language
- JavaScript, with TypeScript used for the MCP server source
- Runtime
- Node.js 18 or newer
- Interfaces
- Claude Code browsing skill with CLI commands, plus an MCP server exposing one browser tool
- Installation surface
- Claude Code marketplace plugin, NPX directly from GitHub, or local clone and build
- Platform support
- macOS, Linux, Windows, WSL2, and container guidance are documented
- Dependencies
- No runtime dependencies are declared; Biome and jsdom are development dependencies
- Testing structure
- The package test script runs linting, bundle freshness validation, and Node test files
Read from README.md, package.json, docs/TESTING-INSTRUCTIONS.md, docs/cdp/INDEX.md, docs/cdp/flatten-mode.md, docs/cdp/target-lifecycle.md, docs/cdp/headless-variants.md, docs/cdp/autoattach-popup-timing.md, docs/cdp/per-session-id-counters.md, docs/cdp/navigation-listener-race.md, docs/superpowers/plans/2026-05-13-dialog-handling.md, docs/superpowers/plans/2026-05-05-post-c5-followup.md, docs/superpowers/plans/2026-05-22-mcp-schema-reshape.md, docs/superpowers/plans/2026-05-06-bug-fixes-after-manual-test.md, test-harness.js.
What it can do
Control browser navigation
URL or navigation commands → Browser navigates to specified pages
Extract data from web pages
Web page content and selectors → Scraped data from DOM elements
Automate form filling
Form data and field selectors → Completed web forms
Click elements on web pages
Element selectors or coordinates → Triggered click events on page elements
Execute JavaScript in browser
JavaScript code → Code execution results in browser context
Take screenshots of web pages
Page or element selectors → Screenshot images
Monitor network requests
Network activity filters → Network request and response data
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.