agents and MCP

Uoink is an MCP server.

Give your local AI agent eyes and ears on the web. Uoink exposes capture, search, hook classification, comment analysis, and entity lookup as local tools.

Copy config Read MCP manifest

install in your client

Copy the config. Then let the agent call tools.

These snippets use placeholder paths. After install, Uoink's setup page generates the exact command for your machine.

Claude Desktop

{
  "mcpServers": {
    "uoink": {
      "command": "%LOCALAPPDATA%\\Uoink\\uoink_mcp.exe",
      "args": []
    }
  }
}

Cursor

{
  "mcpServers": {
    "uoink": {
      "command": "%LOCALAPPDATA%\\Uoink\\uoink_mcp.exe",
      "args": []
    }
  }
}

Cline / Continue

{
  "name": "uoink",
  "transport": "stdio",
  "command": "%LOCALAPPDATA%\\Uoink\\uoink_mcp.exe"
}

13 tools your agent can call

Not a screenshot. Real names, real tools.

Use the new uoink_* names. Legacy yoink_* aliases should remain for migration, but new configs should use Uoink.

uoink_videolocal MCP tool

Extract a structured corpus from one video URL.See /mcp/manifest.json for machine-readable schema.

uoink_playlistlocal MCP tool

Extract up to 10 videos from a playlist URL.See /mcp/manifest.json for machine-readable schema.

get_job_statuslocal MCP tool

Poll a running extraction job.See /mcp/manifest.json for machine-readable schema.

cancel_joblocal MCP tool

Cancel a running extraction job.See /mcp/manifest.json for machine-readable schema.

list_recent_uoinkslocal MCP tool

Return the latest captures from the local library.See /mcp/manifest.json for machine-readable schema.

search_uoinkslocal MCP tool

Run full-text search across transcripts, comments, and titles.See /mcp/manifest.json for machine-readable schema.

get_uoink_corpuslocal MCP tool

Fetch one saved corpus as markdown.See /mcp/manifest.json for machine-readable schema.

analyze_commentslocal MCP tool

Cluster comment themes, products, and disagreements.See /mcp/manifest.json for machine-readable schema.

classify_hooklocal MCP tool

Classify an opening into the nine Hook Type categories.See /mcp/manifest.json for machine-readable schema.

get_taxonomylocal MCP tool

Return the Hook Type taxonomy.See /mcp/manifest.json for machine-readable schema.

get_citation_maplocal MCP tool

Map corpus slugs to source URLs and timestamps.See /mcp/manifest.json for machine-readable schema.

get_uoink_healthlocal MCP tool

Check helper status and diagnostics.See /mcp/manifest.json for machine-readable schema.

find_mentionslocal MCP tool

Find every video or podcast that mentions an entity.See /mcp/manifest.json for machine-readable schema.

Open machine-readable MCP page ->

composing Uoink with other tools

Ask Cursor to research the web video for real.

prompt

Uoink these three competitor videos, classify the hooks, then write a short doc comparing pacing patterns.

result

The agent extracts each video, polls the jobs, fetches the corpora, runs hook classification, then writes the comparison into your repo or notes folder.

tool sequence

uoink_video(url_1)
uoink_video(url_2)
uoink_video(url_3)
get_job_status(job_id)
get_uoink_corpus(corpus_id)
classify_hook(corpus_id)
search_uoinks("pacing")

stdio first

Supported transport is stdio.

The local helper defaults to stdio because that is the path Claude Desktop, Cursor, Cline, and Continue expect. HTTP JSON-RPC at 127.0.0.1:5179/mcp is experimental for local integrations.

Install Uoink -> GitHub ->

agent FAQ

The integration questions first.

What is MCP and how does Uoink use it?

Model Context Protocol lets an AI client call local tools. Uoink exposes a local stdio MCP server so Claude Desktop, Cursor, Cline, Continue, and similar clients can capture and search your library.

Does the agent need the clipboard flow?

No. The agent can call uoink_video, poll get_job_status, then read the corpus with get_uoink_corpus.

Is HTTP transport supported?

Stdio is the supported path. HTTP JSON-RPC on 127.0.0.1:5179/mcp is experimental for local integrations.