Quickstart

This quickstart walks through the Claude Code MCP server setup for Vidmatic, then your first recording. For every tool the server exposes, see the MCP server reference. For what the whole loop looks like from a developer's seat, see screen recording for Claude Code.

Before you start

• A Vidmatic account.
• Claude Code or the Cursor desktop app.
• A repository your agent can already open GitHub issues in.
• Optional: the Vidmatic Chrome extension, if you want to draw on the page while you record.

You do not need an API key.

1. Claude Code MCP server setup

In Claude Code, run this once per machine:

Claude Code adds Vidmatic and lists it as "Needs authentication". Run /mcp, choose vidmatic, then choose Authenticate. Your browser opens a Vidmatic page where you choose an access level. The tools appear in the same session, with no restart.

Using Cursor? Follow the Cursor steps on the MCP server page.

2. Record something

Open the page that is wrong and record it. Either:

• start a recording from the Chrome extension, talk, and draw on the page with the pen, or
• press New Recording in Vidmatic and pick a screen or window. There is no pen in this mode, so say out loud which element you mean.

Recording your screen covers both.

3. Hand it to your agent

When the recording finishes, choose Give to your agent, copy the prompt and paste it into Claude Code. The prompt is:

The agent waits for the transcript if the recording is new, reads what you said and what you drew, finds where each ask lands in your code, and files one GitHub issue. Then it starts the work, and you can follow each stage on Agent Workstreams.

Check the whole pipeline

Run /mcp__vidmatic__record --init once. It records a short test demo, publishes it, prints its public link and puts it in your Inbox. If you see it there, everything is connected. Delete it afterwards if you do not want the public link to stay up.

Remove Vidmatic