Model Context Protocol server

Every Uoink tool is available to your AI: 91 tools over stdio and 91 in the local HTTP registry for capture, search, citations, and writing.

Install Uoink Back to features

Setup page displaying generated MCP command

what it does

The user-facing part.

Give your AI tools access to video data. The local helper runs an MCP server that exposes the local corpus to your AI clients. Your agent can capture videos, query your library, analyze comments, and pull citations without clipboard steps. The server runs on localhost, keeping your research database private.

getting started

Run the loop.

Install the Uoink helper on your machine.
Open the setup page to get the command path.
Add the server to your AI client configuration.
Start the client and verify the connection.
Ask the agent to capture a video to test the setup.

behind the scenes

How Uoink handles it.

Stdio runs the bundled Python against uoink_mcp.py. The separate HTTP JSON-RPC registry is at http://127.0.0.1:5179/mcp/v1 and uses X-Uoink-Token from the local install. HTTP and stdio advertise the same flattened tool schemas; the server still checks the exact schema for every call. Claude Desktop, Claude Code, and Cursor are the tested clients.

mcp reference

Agent-readable surface.

uoink_videosearch_uoinksget_uoink_corpus

{
  "name": "uoink_video",
  "arguments": {
    "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
  }
}

Full server metadata lives at /mcp and /mcp/manifest.json.

try it now

Run it on your next source.

Connect Uoink to your AI clients to enable video and transcript tools.

Install Uoink