A YouTube MCP server gives an AI assistant tools for a specific YouTube workflow. There is no single capability set behind the name. A transcript server may read public caption tracks but have no account login or upload tool; an account connector may use OAuth and expose entirely different operations.
For creators and small businesses, start with the result: a sourced video summary, a table of your channel’s uploads, or a draft upload package. Do not choose a server by its number of tools. This guide checks public documentation on November 12, 2026; it does not claim a live account connection or upload test.
Choose by data and permissions
| Goal | Connection needed | Boundary to verify |
|---|---|---|
| Summarize a public video | A transcript-capable server | Caption availability and timestamp coverage; no guarantee for every public video |
| Read public titles, views and comments | YouTube Data API access, commonly an API key | Quota, pagination and public-data scope |
| Read private account data or modify a channel | OAuth with the appropriate scopes and tools | The actual account/channel and permitted action |
| Study audience retention or revenue | The relevant analytics service or an authorized export | Public video statistics are not private YouTube Analytics |
A concrete community option for Claude Desktop
Promethic-Labs/youtube-mcp is a community server with a Desktop extension. Its documented setup is:
- Review the repository and download
YouTube.mcpbfrom its linked GitHub release. - Open the file with Claude Desktop and review the installation details.
- For transcripts, its implementation does not require an API key. For public search and metadata tools, add a Google project’s YouTube Data API key in the extension settings.
- Start a new conversation with the connector enabled. First list the available tools and read one public video you know.
The maintainer states the Desktop extension uses the app’s runtime; its separate source/CLI setup needs Node and a build. Do not put the extension file into a remote-connector URL field. The official Desktop extension guide explains that local installation surface.
This example is a way to read public data, not a recommendation to give it channel-write access. Its tool list includes transcripts, video/channel metadata, comments and playlists; a public playlist-reading tool does not imply a playlist-editing tool.
Your first research task should preserve the evidence
Read these three video URLs. Return a table with the video URL, publication date, main claims and supporting transcript timestamps. Mark missing captions and gaps. Suggest three original angles for our next product tutorial, separating source facts from your suggestions. Do not post comments or upload anything.
This is an example request, not a test result. Check that each timestamp exists and supports the claim. A title and description are not a transcript; if only those were available, the report should say so. If comments are used, preserve the sample size and sorting method instead of calling a few top comments “what the audience thinks.”
For channel comparisons, control for age. A video published yesterday has had less time to accumulate views than one published last year. Keep publication dates, comparable windows and the retrieval time; do not conclude that a title style caused performance from raw view counts alone.
Transcript, API and quota limits that often get confused
Google’s official caption download endpoint requires permission to edit the video. That is different from the community server’s public-player caption retrieval. The latter can fail when a track is missing, a video is restricted, or YouTube changes its internal endpoints. “No API key for transcripts” is not official unrestricted caption access.
For the official API, an API key identifies a project for public-data calls; OAuth is needed for private user data and changes. Signing in to a channel does not create a transcript that does not exist.
Quota guidance can age quickly. The community README contains an older search-cost table; Google’s current quota calculator separates search and upload buckets from the other endpoints. Use the current Google documentation and your Cloud console, not a blanket “all calls share 10,000 units” assumption. Pagination also consumes calls; start with a bounded list instead of crawling a whole channel.
Using the YouTube connector in Orkas
Orkas’s connector catalog includes YouTube. Open Connectors, find YouTube, complete the offered authorization and confirm its state. Select it for the task, ask for the available operations, and make the first request a small read. Actual tools depend on authorization and the connected service; the catalog description is not proof that every caption, metric or upload will work.
A useful progression is research → outline → script → video draft → reviewed upload package. Give the writing or VideoStudio agent the sourced research and your approved product facts. Keep the final title, description, thumbnail and visibility choice together for review. This article does not establish an automatic publish flow or a successful Orkas YouTube connection.
For the editing step, see Claude video editing skills; for a wider production task, see video production in Orkas.
Diagnose the failure before replacing the tool
| Symptom | Useful check |
|---|---|
| No transcript | Check captions in the player and which track/language the tool requested; report unavailable evidence. |
| Public search fails | Check API enablement, key restrictions and the project’s current quota. |
| Wrong channel | Confirm the authorized account and channel identity before any private read or write. |
| Numbers disagree | Compare the same video, metric definition, period and retrieval time; allow for update delays. |
| Upload or edit tool missing | Inspect the server’s real tool list and OAuth scopes. Do not assume a read connector includes writes. |
FAQ
Is every YouTube MCP server an official Google server?
No. The server used in this guide is a community project. Google’s Data API is official; a wrapper around it is not automatically a Google product.
Can it read every public video’s subtitles?
No. Captions may be missing, restricted or inaccessible to that retrieval method. The official download API additionally requires edit permission.
Does an API key let the agent upload videos?
No. Private data and modifying operations require appropriate OAuth authorization and an implementation that exposes the action.
Can public view counts replace retention data?
No. Public statistics do not establish where viewers leave or how a video earned revenue. Use authorized analytics or an export for those questions.
Sources and scope
Community server and tool list; Google’s caption download, authorization and quota documents. Checked November 12, 2026. Setup is documented, not live-tested with a YouTube account; no transcript retrieval or upload is presented as an experiment here.
Reddit MCP for customer discussion research →