Analysis updated 2026-05-18
Have an AI assistant debug a CSS layout problem on a live web page just by describing the issue.
Automatically write Cypress end to end tests by walking an assistant through a sign up flow in plain language.
Crawl a staging site with an assistant and generate a Markdown document describing every page it finds.
Run open ended exploratory testing where the assistant tries edge case inputs and reports what breaks.
| atagon-gmbh/kogiqa-mcp | foundzigu/smsbazaar | mrdoob/draco.js | |
|---|---|---|---|
| Stars | 151 | 150 | 153 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | — | easy |
| Complexity | 2/5 | — | 2/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 20 or newer and an MCP-compatible client like Claude Code, Cursor, or VS Code.
kogiQA MCP Web Browser is a server that follows the Model Context Protocol, or MCP, a standard that lets AI assistants call external tools. This particular server gives an AI assistant control of a web browser, so it can open pages, click things, and read what is on screen, all described through plain natural language instructions rather than screenshots or coordinates. Because it works without needing to point at specific page elements by name, the README says it saves time and reduces how much text the assistant has to process compared to tools that require selectors. The README lists several ways people use it: finding and describing style problems on a page, automatically fixing errors that show up in the browser console, exploring an app to write documentation of what each page does, generating end to end tests such as Cypress tests from a walkthrough of a user flow, and doing open ended exploratory testing where the assistant tries different inputs and reports anything broken. Installing it is a single command, npx kogiqa-mcp@latest, and the README gives specific setup steps for Claude Code, VS Code, and Cursor, plus a generic JSON configuration block for any other MCP compatible client. Once configured, a user talks to their AI assistant normally, for example asking it to open a local site, log in with test credentials, add an item to a cart, and check that a cart counter updates correctly. Other example prompts include checking why a pricing page element overflows its container at a certain screen width, or crawling a staging site up to two levels deep and writing a document describing every page found. Using it requires Node.js version 20 or newer and an MCP capable client such as VS Code, Cursor, Windsurf, Claude Desktop, Goose, or Junie. The README does not state what license the project is released under.
An MCP server that lets an AI assistant control a real web browser using plain language instead of screenshots or selectors.
Mainly JavaScript. The stack also includes JavaScript, Node.js, MCP.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.