MCP Server

Ask Claude for a demo video

QuickPeek ships an MCP server, so an agent can record a narrated demo of a web app directly. One command to install. The first five videos over MCP need no account, no email and no card.

Install

Claude Code:

bash

claude mcp add quickpeek -- npx -y --ignore-scripts --package=@spark-apps/quickpeek quickpeek-mcp

Any other MCP client, including Claude Desktop, Cursor and Windsurf. Add this to its config file:

json

{
  "mcpServers": {
    "quickpeek": {
      "command": "npx",
      "args": ["-y", "--ignore-scripts", "--package=@spark-apps/quickpeek", "quickpeek-mcp"]
    }
  }
}

Registry id: co.quickpeek/quickpeek
npm package: @spark-apps/quickpeek
Transport: stdio
Requires: Node 18+ and ffmpeg

Listed on the official MCP registry. There is no API key to get: the AI runs through QuickPeek's hosted relay.

Then just ask

Once it is installed, the request is plain language. The recording takes minutes, not seconds, because a real browser session is filmed.

“Record a demo video of localhost:3000 and narrate it.”

The finished MP4 lands in the project it demos, beside the editable plan.json that produced it. Nothing is uploaded.

The nine tools

Tool	What it does
generate_demo	Plans a walkthrough, drives a real browser, narrates and captions the result.
generate_demo_from_script	You supply the narration lines, it maps each one to a UI action.
list_demos	Every demo already recorded for a URL, newest first.
get_demo_config	Canvas size, captions, timing and voice: what the next demo will look like.
list_voices	Narration voices in 60+ languages. Free, no API key.
open_in_editor	Opens a recorded demo in the plan editor.
check_account	Your plan, and whether the next demo carries a watermark.
start_auth	Opens a visible browser so you can sign into the app being demoed.
finish_auth	Saves that session, so the demo starts inside the app, not on its login page.

Your model does the writing

The server never calls an AI of its own. It hands the planning and narration prompts back to the model that called it and takes the answer as a parameter. Two consequences worth knowing: the demo is written by whatever model you are already talking to rather than a cheaper one behind our relay, and recording over MCP spends no QuickPeek AI quota at all.

Prefer a terminal?

Same product, no MCP client. One difference worth knowing: the CLI has no model of its own, so it reaches our hosted planner and asks for an email on the first run to meter it. No card, but that confirmation link does need clicking before the first video.

bash

npx @spark-apps/quickpeek https://your-app.com

Runs on your machine. Nothing about your app is uploaded.