Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@videoseek-mcp find moments where someone is cooking pasta"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Find anything in any video. Semantic video search, video Q&A, persistent memory, and social media import — all as MCP tools for your AI agent.
Give your AI agent (Claude, Cursor, Windsurf, etc.) the ability to:
Search videos by meaning — "find the moment where someone opens the gift" across your entire library
Ask questions about videos — multi-turn conversations with one video or your whole collection
Remember everything — persistent text + video memory that never forgets (Memory Augmented Generation)
Import from social media — pull and index content from TikTok, YouTube, Instagram, and 15+ platforms
Analyze on-the-fly — describe any video or image from URL without uploading
Transcribe — get visual scene descriptions or spoken word transcripts
Related MCP server: klaket-mcp
Sign up at memories.ai and grab your key from the API keys page. Free tier includes 100 credits/month.
git clone https://github.com/kennyzheng-builds/videoseek-mcp.git
cd videoseek-mcp
npm install --include=dev
npm run buildAdd to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
{
"mcpServers": {
"videoseek": {
"command": "node",
"args": ["/path/to/videoseek-mcp/build/index.js"],
"env": {
"MEMORIES_API_KEY": "your-api-key"
}
}
}
}Add to .cursor/mcp.json:
{
"mcpServers": {
"videoseek": {
"command": "node",
"args": ["/path/to/videoseek-mcp/build/index.js"],
"env": {
"MEMORIES_API_KEY": "your-api-key"
}
}
}
}export MEMORIES_API_KEY="your-api-key"
claude mcp add videoseek -- node /path/to/videoseek-mcp/build/index.jsTool | What it does |
| Upload and index a video from URL |
| List all indexed videos |
| Check processing status |
| Remove videos from your library |
| Get visual or audio transcription |
Tool | What it does |
| Semantic search across your private video library |
| Search TikTok, YouTube, Instagram by meaning |
| Find when something was said in a video |
Tool | What it does |
| Ask questions about specific videos (multi-turn) |
| Ask questions across your entire library + memories |
Tool | What it does |
| Store text with semantic indexing |
| Find relevant memories by meaning |
| List all stored memories |
Tool | What it does |
| Analyze a video from URL (no upload needed) |
| Analyze an image from URL |
Tool | What it does |
| Import from any social media URL |
| Import by hashtag from TikTok/YouTube/Instagram |
| Import from a creator's profile |
Resources: memories://videos (your library), memories://memories (your stored knowledge)
Prompt templates: analyze-video, social-media-research, build-knowledge-base
"Find a moment"
You: Find all moments where someone is cooking pasta
Agent uses
search_videos→ returns timestamped video segments
"Understand a video"
You: What's happening in this video? [URL]
Agent uses
caption_video→ instant analysis without uploading
"Research TikTok"
You: Research AI trends on TikTok
Agent uses
search_public→import_from_url→chat_personalfor insights
"Build knowledge"
You: Summarize all my meeting recordings
Agent uses
list_videos→get_transcription→add_memoryfor key takeaways
Feature | videoseek-mcp | TwelveLabs MCP |
Persistent memory (text + video) | Yes | No |
Social media import (15+ platforms) | Yes | No |
Memory Augmented Generation | Yes | No |
Free tier | 100 credits/month | 10 hours |
Multi-video Q&A | Yes | Yes |
Semantic search | Yes | Yes |
Variable | Required | Default | Description |
| Yes | — | API key from memories.ai |
| No |
| Namespace for multi-tenant isolation |
npm install --include=dev
npm run build # compile TypeScript
npm run dev # watch mode
npx @modelcontextprotocol/inspector node build/index.js # test interactivelyvideoseek-mcp/
├── src/
│ ├── index.ts # MCP server — 18 tools, 2 resources, 3 prompts
│ └── client.ts # Video understanding API client
├── build/ # Compiled JS (generated)
├── package.json
└── tsconfig.jsonTypeScript + Zod schemas + stdio transport. No dependencies beyond MCP SDK.
MIT
Store a text memory with semantic indexing. Memories are searchable and integrated into personal chat responses via Memory Augmented Generation (MAG). Use this to store notes, meeting summaries, insights, or any text knowledge.
Inputs: content (required), unique_id, tags.
Analyze an image from URL using AI vision. Describe scenes, identify objects, read text, and answer questions about the image.
Inputs: image_url (required), prompt (required), system_prompt.
Analyze a video from URL without uploading to your library. Uses the Large Visual Memory Model (LVMM) to understand and describe video content. Supports optional reasoning mode for complex analysis tasks.
Inputs: video_url (required), prompt (required), system_prompt, thinking.
Ask questions across your entire video and memory library. This is the most powerful query tool — it combines video understanding with text memories (MAG: Memory Augmented Generation) to provide comprehensive answers. Recommended for general questions about your content.
Inputs: prompt (required), unique_id, session_id.
Ask questions about specific videos using natural language. The AI analyzes the video content and provides detailed answers. Supports multi-turn conversations via session_id.
Inputs: video_nos (required), prompt (required), unique_id, session_id.
Delete one or more videos from your VideoSeek library by their video numbers.
Inputs: video_nos (required), unique_id.
Retrieve the transcription of a video. Supports both video transcription (visual scene descriptions) and audio transcription (spoken words).
Inputs: video_no (required), type, unique_id.
Check the processing status of a video by its task ID. Returns video numbers and their current status (PARSE=ready, UNPARSE=processing, FAIL=failed).
Inputs: task_id (required), unique_id.
Import videos from a social media creator's profile URL. Indexes their recent videos into your library for analysis.
Inputs: creator_url (required), count, unique_id.
Import videos from a social media platform by hashtag. Automatically imports and indexes matching videos into your library.
Inputs: hashtag (required), platform, count, unique_id.
Import a video from a social media URL (TikTok, YouTube, Instagram, Twitter, etc.) directly into your VideoSeek library for indexing and analysis.
Inputs: url (required), unique_id, tags.
List all stored text memories in the current namespace.
Inputs: unique_id, page, page_size.
List all indexed videos in your VideoSeek library. Supports pagination and filtering by status.
Inputs: unique_id, page, size, status.
Search within a specific video's audio transcripts. Useful for finding when specific topics were discussed in a video.
Inputs: video_no (required), query (required), unique_id.
Semantic search across your stored text memories. Returns the most relevant memories ranked by similarity to your query.
Inputs: query (required), unique_id, page, page_size.
Search for videos on public platforms (TikTok, YouTube, Instagram) using semantic natural language queries. Great for content research and discovery.
Inputs: query (required), platform, top_k.
Semantic search across your private video library using natural language. Find specific moments, scenes, objects, or actions across all your indexed videos. Returns matching video segments with timestamps.
Inputs: query (required), unique_id, top_k, tag, +1 more.
Upload and index a video from a URL into your VideoSeek library. Returns a videoNo for tracking. The video will be processed asynchronously — use get_video_status to check when it's ready.
Inputs: url (required), unique_id, callback, tags.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 18 tools
All 18 tools have clearly distinct purposes, covering memory management, video operations, imports, and searches without overlap. Even similar tools like chat_personal and chat_with_video target different scopes (library-wide vs per-video).
All tool names follow a consistent verb_noun pattern using underscores (e.g., add_memory, caption_image, delete_videos). No mixing of conventions or camelCase.
18 tools is well-scoped for a video and memory management server. The count covers all necessary operations without being excessive or sparse.
The tool set provides comprehensive CRUD-like operations for videos (upload, list, delete, get status, transcribe, caption), search (semantic, audio, public), chat, imports, and memory. No obvious gaps for the intended domain.
Transcripts, summaries, chapters and timestamped answers for video links, for AI agents.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Persistent memory for AI agents. Search, store, and recall across sessions.
Video knowledge base for agents: search your library's transcripts, keyframes and on-screen text.
