memvid

The memvid MCP server provides a Streamable-HTTP interface that leverages memvid to encode text data into a video index. It exposes two primary actions you can perform: add_chunks and search. The add_chunks action appends new text chunks to the in-memory video index; note that each time chunks are added, the underlying memory.mp4 is reset. The search action queries the index and returns the top matching chunks (default top_k is 5, adjustable via the top_k parameter). Clients connect to the server using a configured mcp-config.json file (as shown in the example) and interact with the HTTP endpoints defined by the streamable-http interface. Use memvid to encode semantic representations of text into a timeline-based video format, enabling fast lookups through semantic search.

Typical workflow: start the Python server, then use the MCP client configuration to connect and issue add_chunks with your data, followed by search calls to retrieve relevant chunks. The server supports configuring a non-default port via environment variables (for example, PORT=3002 as shown in the README).

Prerequisites:

• Python 3.11 (or compatible 3.x environment)
• Python virtual environment support
• access to the project repository with a requirements.txt file

Step-by-step:

1. Create and activate a virtual environment python3.11 -m venv my_env source my_env/bin/activate

2. Install dependencies pip install -r requirements.txt

3. Run the server (default port 3000) python server.py

4. (Optional) Run on a custom port PORT=3002 python server.py

5. Verify the server is accessible and the MCP config example can be used to connect

   • You can test by sending HTTP requests to http://localhost:3000 (or your configured port) using your preferred MCP client.

Note: If you modify environment variables or ports, ensure the mcp-config.json used by clients points to the correct URL.

Tips and common considerations:

• The add_chunks action resets memory.mp4 each time new chunks are added. If incremental updates are required, consider batching your input or checking for an updated memvid implementation.
• The search action defaults to returning 5 results but supports adjusting top_k via the request payload parameters.
• If you change the server port (via PORT or other means), update the mcp-config.json so clients connect to the correct URL.
• Ensure your Python virtual environment has the correct Python version (Python 3.11 as per the setup instructions).
• If you encounter connectivity issues, verify that the server is running and listening on the expected port, and that firewalls or proxies aren’t blocking requests to http://localhost:<port>.