Kaboom MCP
An MCP server and Chrome extension feed an agent console logs, network errors, exceptions, action recordings and page audits.
- 67 ★ on the repository
- Last commit
Broad access
Why this level
- The extension hands the agent network response bodies, cookies and page content, and the agent can click and type in the tracked tab.
- Installation registers autostart and changes MCP client configuration.
Install
- Works in
Manual install
curl -sSL https://raw.githubusercontent.com/brennhill/Kaboom-Browser-AI-Devtools-MCP/STABLE/scripts/install.sh | bashThe README command for macOS and Linux. For Windows the README gives a PowerShell `irm ... install.ps1 | iex` command. Read the script before running it.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The extension hands the agent network response bodies, cookies and page content, and the agent can click and type in the tracked tab.
- Installation registers autostart and changes MCP client configuration.
This is third-party code. Review the repository files before installing.
What it does
Kaboom consists of a local Go binary and a Chrome extension (Manifest V3). Together they pass what happens in the browser to a coding agent over MCP: console messages, failed requests with response bodies, exceptions with stack traces, WebSocket events, recordings of clicks, typing and scrolling, and Web Vitals. The agent can query the DOM by CSS selector, run a WCAG accessibility audit with SARIF export and a security audit (credentials, personal data, headers, cookies, third-party scripts), and drive the page: click, type, select values, upload files and navigate. From recorded actions Kaboom generates Playwright tests and reproduction scripts. No Chrome debug port is required: an ordinary extension does the work. The server listens on `127.0.0.1` only, authorisation headers are stripped, and there is no cloud or account. The install script downloads the binary, places the extension in `~/KaboomAgenticDevtoolExtension` and configures the MCP clients it detects: Claude Code, Cursor, Windsurf, Zed.
Who it is for. Web developers who debug an app in Chrome with Claude Code, Cursor or Windsurf and want to give the agent live browser data.
Good fit when
- A page crashes with an exception and the agent needs the stack and recent network errors.
- You need Playwright tests built from actions you performed in the browser.
- Before a release the current page needs an accessibility or security audit.
- A Web Vitals regression appeared after an edit and the agent must compare the metrics.
Not a fit when
- You work outside Chrome or in a browser without Manifest V3 extensions.
- Running remote install scripts on the machine is forbidden.
- You need to stay free of AGPL-3.0: the licence requires opening derived code.
Example request
Open my page, reproduce the save error on the form and show the console and failed network requests with response bodies.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Take code from the `STABLE` branch: `UNSTABLE` makes no promises. The extension is loaded manually through `chrome://extensions` in developer mode, then you click Track This Tab in the popup. Installation uses a `curl | bash` script, so read it before running, and locked-down environments have strict mode `KABOOM_INSTALL_STRICT=1`. Anonymous usage statistics are collected and the opt-out setting is described in the README. The licence is AGPL-3.0.
How to disable. Run `uninstall.sh` from `STABLE/scripts` with `--yes`: it removes the binary, extension files, autostart and MCP client settings. Then remove the extension in `chrome://extensions`.
FAQ
How do I install Kaboom?
Run the install script from the `STABLE` branch, then load `~/KaboomAgenticDevtoolExtension` as an unpacked extension in `chrome://extensions` and restart the AI tool.
Is a Chrome debug port needed?
No. Kaboom works as an ordinary extension, you launch the browser as usual and the sandbox stays on.
How do I configure it manually?
For Claude Code run `claude mcp add-json --scope user kaboom-browser-devtools` with the path to the `~/.kaboom/bin/kaboom-agentic-browser` binary. Other clients take the same path in their MCP configuration.
Where does the data go?
Logs, network and actions stay on your computer. Only anonymous usage statistics leave it, and they can be turned off.
Which clients are supported?
The README names Claude Code, Cursor, Windsurf, Zed, Claude Desktop and VS Code with Continue.
Discussion
No questions or reviews yet.
Related
MCP serverskogiQA MCP gives an agent a browser that finds page elements without selectors, for debugging web apps and writing e2e tests
An MCP server from Mozilla: an agent drives Firefox through WebDriver BiDi, reads the console and network, takes page snapshots and profiles loading
Electron debugging MCP server: 111 tools let an agent see the window, click buttons, read console and network and audit security.
MCP Accessibility Scanner runs axe checks, exercises keyboard focus and saves accessibility reports for web pages
BrowserStack MCP Server: testing on real devices
BrowserStack MCP Server
The official BrowserStack MCP server: real devices, automated tests and test cases driven by an agent from your editor
WebdriverIO MCP: browser and mobile app automation
webdriverio-mcp
The official MCP from the WebdriverIO team drives Chrome, Firefox, Edge, Safari, Electron, iOS and Android through one tool set
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Kaboom (this card) | no | not required | 67 | |
| Chrome DevTools MCP | yes | not required | 55,126 | the official Chrome DevTools team MCP over the debug port; Kaboom works through an extension and adds audits and test generation |
| BrowserTools MCP | no | not required | 7,342 | console, network, screenshots and Lighthouse from an open Chrome; Kaboom adds action recording, page control and Playwright tests |
README in short
The repository keeps two branches: `STABLE` for use and `UNSTABLE` for development. Documentation, a quick start and the feature list are published on the project site; the licence is AGPL-3.0, and logos and artwork carry a separate licence.