# Amendor

> Amendor lets the people you build software for request changes directly on your live site. Each request lands on your board, your coding agent builds it on a branch and opens a pull request, your host builds a deploy preview, the requester approves, and you merge. It replaces the back-and-forth of emails and screenshots between a developer and the non-technical people they build for.

Who it is for: developers, freelancers, agencies, and studios who ship software for clients, stakeholders, or non-technical users who give feedback on the UI.

How the loop works:
1. A user points at any element on your live site and describes the change they want.
2. The request lands on the developer's board with the page URL, the exact element (CSS selector + outerHTML), and an optional screenshot.
3. The developer's coding agent (Claude Code, Cursor, Cline, Codex, Lovable, any MCP client) pulls the request, builds it on a branch, and opens a pull request.
4. The host (Netlify, Vercel, wherever you deploy) builds a deploy preview from the PR.
5. The requester previews the change and approves it, or asks for revisions.
6. The developer merges and ships.

Install the widget: add one script tag to the site you ship.
`<script src="https://amendor.site/widget.js" data-project="YOUR-PROJECT-KEY"></script>`
The trigger can be the built-in floating button or any element on your page.

Connect your coding agent (Model Context Protocol):
- Local agents (stdio): npm package `amendor-mcp`. Install command with your token is shown on the Amendor settings page.
- Cloud agents (remote Streamable HTTP): connect to `https://amendor.site/mcp` with header `Authorization: Bearer <connector token from settings>`.
- Tools: list_change_requests, get_change_request, set_preview_url, update_status (remote) plus start_build, publish_preview, ship_change (local stdio).

What makes it different:
- It never touches your stack. The agent builds on a branch and opens a normal PR. You stay in control of what merges.
- Requests carry the exact element and page, so the agent has real context, not a vague description.
- The non-technical requester gets a real preview and a say, without needing a GitHub account or your codebase.

Links:
- Home: https://amendor.site
- How it works (demo): https://amendor.site/demo
- About: https://amendor.site/about
- MCP connector (npm): https://www.npmjs.com/package/amendor-mcp
- MCP connector (source): https://github.com/deslay1/amendor-mcp

Contact: https://amendor.site