Building a Video Research Tool with Claude Code

Developers watch a lot of video content — conference talks, library tutorials, recorded architecture reviews. The problem is that none of it is searchable or referenceable from your terminal. When you’re deep in a Claude Code session and you need context from a video, you’d normally have to break out, watch or skim the video, and come back. AgentPatch gives Claude Code the ability to pull that content directly.

Why This Matters

When you’re writing code, writing docs, or researching a technical decision, video content is often where the good explanations live. A library author’s conference talk. A recorded team discussion about an API design. An official product demo that explains behavior the docs gloss over.

Claude Code can use the AgentPatch YouTube Transcript tool to fetch that content mid-session. You get the transcript with timestamps, and Claude Code can reason over it — extracting specific explanations, finding references, or using it to inform something it’s writing.

Setup

The AgentPatch CLI is designed for AI agents to use via shell access. Install it, and your agent can discover and invoke any tool on the marketplace.

Install (zero dependencies, Python 3.10+):

pip install agentpatch

Set your API key:

export AGENTPATCH_API_KEY=your_api_key

Example commands your agent will use:

ap search "web search"
ap run google-search --input '{"query": "test"}'

Get your API key from the AgentPatch dashboard.

Install the AgentPatch skill — it teaches Claude Code when to use AgentPatch and how to use the CLI:

/plugin marketplace add fullthom/agentpatch-claude-skill
/plugin install agentpatch@agentpatch

MCP Server (Alternative)

If you prefer raw MCP tool access instead of the skill:

claude mcp add -s user --transport http agentpatch https://agentpatch.ai/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Replace YOUR_API_KEY with your actual key from the AgentPatch dashboard.

Example

You’re writing a technical design doc and you know there’s a recorded discussion from your team where someone explained the tradeoffs. You tell Claude Code:

“Pull the transcript from this recording and find the part where we discussed caching strategy: https://youtube.com/watch?v=internal-recording”

Claude Code fetches the transcript through AgentPatch and surfaces the relevant section. You paste the timestamp reference into your doc.

Another scenario: you’re evaluating an open-source library and find a demo video linked from the README.

“Get the transcript from this video and summarize what the author says about the configuration options.”

Claude Code retrieves the transcript and gives you the summary. You didn’t open a browser. You didn’t watch a 20-minute demo to find one specific answer.

For multi-video research, Claude Code can handle several at once:

“Here are two talks on this topic. Fetch both transcripts and compare how each author approaches error handling.”

Wrapping Up

Video research is one use case. Once AgentPatch is in your Claude Code config, the same connection gives you access to web search, email, image generation, and the rest of the marketplace — no additional setup needed. Visit agentpatch.ai to get started.

← Back to blog