memories-ai-mcp

MCP (Model Context Protocol) server for the Memories.ai video analysis API. Enables Claude Desktop to upload, transcribe, search, summarize, and chat with video content from TikTok, YouTube, and Instagram.

Features

• Upload videos from URLs or creator profiles (TikTok, YouTube, Instagram)

• Transcribe audio (spoken words) and video (visual scene descriptions)

• Summarize videos by chapter or topic

• Semantic search across private and public video libraries

• Keyword search across audio transcripts

• Chat with videos — ask questions, compare content, generate captions

Related MCP server: screen-memory-mcp

Getting an API Key

1. Sign up at memories.ai

2. Navigate to your account settings / API section

3. Generate an API key

4. See the API docs for details

Setup

Build

npm install
npm run build

Environment Variable

Set your API key:

export MEMORIES_AI_API_KEY="your-api-key-here"

Claude Desktop Configuration

Add this to your Claude Desktop config file (claude_desktop_config.json):

{
  "mcpServers": {
    "memories-ai": {
      "command": "node",
      "args": ["/absolute/path/to/memories-ai-mcp/dist/index.js"],
      "env": {
        "MEMORIES_AI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Available Tools

Typical Workflow

1. Upload a video with upload_video_from_url

2. Poll with get_task_status until status is PARSE

3. Analyze using transcription, summary, search, or chat tools

Important Notes

• video_no ≠ task_id: Upload returns a task_id. Use get_task_status to get the video_no needed by all other tools.

• Async processing: Videos must reach PARSE status before analysis tools work. Error code 0001 means still processing.

• Platform batching: Multiple URLs in one upload must all be from the same platform.

• Public library is permanent: Videos uploaded to the public library cannot be deleted.