webmcp-kit

# WebMCP Kit Make your website or web app agent-ready. WebMCP Kit is a coding-agent plugin that adds [WebMCP](https://github.com/webmachinelearning/webmcp) tools to your site or application, so browser agents can act through your app's own logic instead of scraping the page. It reads your repo, proposes a tool plan, **waits for your approval**, implements the tools with [`@nekuda/webmcp-sdk`](https://www.npmjs.com/package/@nekuda/webmcp-sdk), and verifies them in a real browser. Everything runs locally — **your code never leaves your machine**; there is no hosted scanner. Works with **Claude Code**, **Codex**, and **Cursor**. ## Install **Claude Code** ``` /plugin marketplace add nekuda-ai/webmcp-kit /plugin install webmcp-kit@nekuda ``` **Codex** ```sh codex plugin marketplace add nekuda-ai/webmcp-kit codex plugin add webmcp-kit ``` Or run `/plugins` inside Codex and install it from the browser there. On Codex, installing also registers the plugin's lifecycle hooks (`PreToolUse`/`Stop`) — they only ferry feedback from the live review UI back to the agent, and you can inspect them anytime with `/hooks`. **Cursor** ```sh npx skills add nekuda-ai/webmcp-kit --skill '*' --agent cursor ``` This installs all three Agent Skills for Cursor: `implement`, `verify`, and `connect-existing-tools`. ## Package for the OpenAI Platform Build the skills-only ZIP accepted by the OpenAI Platform plugin uploader: ```sh python3 scripts/package-openai-skills.py ``` The archive is written to `dist/webmcp-kit-<version>.zip`. It includes the Codex manifest, brand assets, and all three skills, and leaves out repository files and the lifecycle hooks used by the full marketplace install. ## Prerequisites - [Claude Code](https://claude.com/claude-code), [Codex](https://developers.openai.com/codex), or [Cursor](https://cursor.com/) - A website or web app you can run locally — any stack, the plugin adapts - [Bun](https://bun.sh) *(optional)* — powers the live review UI; without it you review the plan in chat - Chrome 150+ with the WebMCP flag enabled (`chrome://flags`) — used to verify tools in a real browser ## Use Open your repo in your agent and ask it to make your website or web app agent-ready (Claude Code: `/webmcp-kit:implement`; Cursor: `/implement`). Then: 1. **Plan** — it maps your routes, forms, and data layer, and proposes a small set of tools. 2. **Approve** — nothing is written until you say yes. 3. **Implement** — typed tools wired to your app's own client logic. 4. **Verify** — each tool is checked in a real browser and reported as **verified**, **failed**, or **could-not-verify**. Failed never ships. Already have WebMCP tools? The `verify` skill (`/webmcp-kit:verify`) checks they register and work. ## License [MIT](LICENSE)