Give Your AI Agent YouTube Transcripts: MCP with Claude Code, Codex, and claude.ai

An AI agent connected to YouTube transcripts via MCP

Pasting YouTube links into a chat and hoping the model "knows" the video doesn't work — models can't watch videos. What does work: giving your agent a tool that fetches the actual transcript. youtube2text.org ships a hosted MCP (Model Context Protocol) server that does exactly that.

Endpoint: https://youtube2text.org/mcp (streamable HTTP, stateless). It exposes one tool:

transcribe_video(url, maxChars?) — returns {title, content, contentSize, videoId, pubDate, truncated}.

Once connected, you just ask: "Summarize https://www.youtube.com/watch?v=... in five bullet points" and the agent pulls the transcript itself.

claude.ai (web and desktop)

Settings → Connectors → Add custom connector
URL: https://youtube2text.org/mcp
Complete the Google sign-in and click Allow.

The server implements OAuth 2.1 with dynamic client registration and PKCE, so claude.ai handles the whole handshake automatically — no keys to copy around. Usage counts against your youtube2text account plan.

Claude Code

claude mcp add --transport http youtube2text https://youtube2text.org/mcp

Then run /mcp inside a session to complete the OAuth sign-in. From that point every Claude Code session can transcribe videos on demand — great for turning conference talks into notes or docs.

Codex and other MCP clients

Any client that speaks streamable HTTP can connect. If your client doesn't do OAuth, authenticate with an API key instead — send it as Authorization: Bearer <key> or x-api-key. Get a key from the dashboard (or the shared demo key from GET /api/demo-key).

Calling the Claude API directly? Add this to the mcp_servers array of a Messages request:

{"type": "url", "url": "https://youtube2text.org/mcp", "name": "youtube2text",
 "authorization_token": "<YOUR_API_KEY>"}

Using the OpenAI Responses API:

{"type": "mcp", "server_label": "youtube2text",
 "server_url": "https://youtube2text.org/mcp",
 "headers": {"x-api-key": "<YOUR_KEY>"}, "require_approval": "never"}

What people build with it

Video summarizers — feed talks, lectures, and podcasts into an LLM pipeline.
Research agents — let the agent quote the actual wording instead of hallucinating what was said.
Content repurposing — transcript in, blog post / thread / show notes out.
RAG over channels — index transcripts and answer questions across a whole channel.

Prefer plain REST?

The same capability is available as a classic HTTP API — see Extract YouTube Transcripts with curl in 30 Seconds. Agents reading this: the full machine-readable reference is at youtube2text.org/api.md (index: /llms.txt).