# FlowScope UX Flow Analyser [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/DavidNgugi/flowscope  
**GitHub Stars:** 2  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/flowscope-ux-flow-analyser

## Description
Turn YouTube product demos into structured UX flow reports, and compare demos side by side.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "flowscope-ux-flow-analyser": {
    "command": "uvx",
    "args": ["--from","flowscope-mcp","flowscope-mcp-install"],
    "env": {
      "ANTHROPIC_API_KEY": "",
      "OPENAI_API_KEY": "",
      "DEEPSEEK_API_KEY": "",
      "OPENAI_COMPATIBLE_API_KEY": "",
      "OPENAI_COMPATIBLE_BASE_URL": "",
      "YTDLP_COOKIES_FROM_BROWSER": "",
      "YTDLP_COOKIE_FILE": ""
    }
  }
}
```

**Requires environment variables:** `ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `DEEPSEEK_API_KEY`, `OPENAI_COMPATIBLE_API_KEY`, `OPENAI_COMPATIBLE_BASE_URL`, `YTDLP_COOKIES_FROM_BROWSER`, `YTDLP_COOKIE_FILE` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What FlowScope UX Flow Analyser does

FlowScope UX Flow Analyser MCP server connects an MCP-capable client to a local REST backend that analyses YouTube product-demo videos. Its pipeline produces a structured account of the product flow, visible screens, and UX patterns. It can also compare the flows found in multiple videos, making it suitable for competitive UX research based on publicly available demos.

The system keeps downloaded videos and derived analysis on the local machine. A single analysis generally takes 2–10 minutes and can make one vision-model request for each distinct screen identified in the video. Results are stored and keyed by YouTube video ID, so checking existing analyses or reading a completed report does not repeat the paid processing stages.

## How FlowScope UX Flow Analyser MCP server works

The MCP package acts as a stateless client of the backend REST API. It does not import the application code and does not require ffmpeg bindings or Whisper installed in the MCP package itself. The backend handles video retrieval, transcription, frame extraction, screen deduplication, narration alignment, and LLM-based synthesis.

Transcription can use available captions or local Whisper. The frame stage requires a vision-capable model. Separate model settings can be assigned to frame analysis, single-video synthesis, and cross-video comparison, with each purpose falling back to the default provider and model when no override is configured.

Bundled skills guide an agent to check the cache before starting an expensive run. A health-check tool reports the state of the backend, ffmpeg, yt-dlp, and the configured LLM provider key.

## Setup and configuration

Start the backend with Python dependencies installed and run it with Uvicorn on port 8000. The repository also documents a Docker Compose setup, with backend data bind-mounted so the database, media, and frame files remain on the host. The frontend is optional for MCP use but can be started separately for a local web interface.

The MCP distribution is available through PyPI as `flowscope-mcp`. Its installer can generate client-specific configuration for clients including Cursor, VS Code, Codex, and Gemini. Claude Code has a repository plugin installation path. The two bundled skills can be copied into `.agents/skills/` or installed with the skills CLI.

Supported LLM provider selections include Anthropic, OpenAI, DeepSeek, and OpenAI-compatible endpoints. Credentials use the corresponding provider key variables. YouTube authentication can be configured with `YTDLP_COOKIES_FROM_BROWSER` for local development or `YTDLP_COOKIE_FILE` when supplying a Netscape-format cookie file to a container.

## Tools and capabilities

FlowScope UX Flow Analyser exposes the analysis pipeline through MCP tools. Supported workflows include:

- Run a UX teardown for one YouTube demo.
- Ask how many steps a signup or onboarding flow contains.
- Compare onboarding or other flows across two videos.
- Retrieve previously analysed videos without paying for another analysis.
- Check backend and analysis prerequisites with `flowscope_health_check`.

## Limitations and notes

The tool is intended for personal competitive UX research. Processing requires access to the YouTube video and may require authenticated yt-dlp cookies if YouTube presents an anti-bot challenge. Vision analysis and LLM synthesis can incur provider charges. Model compatibility matters: text-only models cannot perform the frame stage, and unsupported or missing provider credentials produce configuration errors.

_Full upstream README: https://allmcps.com/mcp/flowscope-ux-flow-analyser/readme_