Editing & Tools

YouTube MCP Servers: What Each Can Do for Your Agent

ChatCut Team
15 min read
YouTube MCP Servers: What Each Can Do for Your Agent
Contents
  1. YouTube MCP servers at a glance
  2. How we chose these servers
  3. Search and public data servers
  4. Transcript servers
  5. Upload and channel management servers
  6. Owner analytics servers
  7. Choose access and Google scopes
  8. Turn your own video into Shorts with ChatCut

A YouTube MCP server gives an AI agent such as Claude, Codex or Cursor tools for YouTube, and the servers split into four jobs: searching public videos, pulling transcripts, uploading to your channel and reading your channel’s analytics. Choose by job. Public research needs at most a Google API key, while uploads and analytics need permission to your Google account, so check the exact scopes first. This guide compares 15 servers on what they do, how they sign in and what they cost, based on their documentation and source code as of November 12, 2026, and ends with turning your own long video into Shorts.

YouTube MCP servers at a glance

ServerJobAccess
ZubeidHendricksSearch, transcripts, channelsAPI key
vidlens-mcpLocal research libraryNo key to start
AnarcystSearch, transcripts, channelsNo key
TranscriptAPIHosted transcripts, searchVendor account
kimtaeyoon83Transcripts, language fallbackNo key
anaisbettsSubtitles via yt-dlpNot documented
jkawamotoTimed transcriptsNo sign-in documented
TranscriptOutHosted batch transcriptsVendor API key
SeanPedersenTranscripts in 5 languagesNot documented
Eli-xirCaptions, local fallbackNo key
youtube-mcp-cliPlain or timed transcriptsNo key
yutuUploads, captions, commentsGoogle OAuth
youtube-studio-mcpUploads, analyticsGoogle OAuth
ZernioPublishing, schedulingVendor + channel
youtube-kitOwner analytics, uploadsGoogle OAuth

“Access” is what the README documents. Self-hosted servers also run on your computer, so you install their dependencies.

How we chose these servers

We looked for YouTube MCP servers on GitHub and vendor sites, read each README and, for the owner tools, the source code that requests Google permissions. We grouped them by job and compared what they return, how they authenticate and what they cost. GitHub stars and last-update dates come from GitHub’s API on November 12, 2026; stars show attention, not quality. We didn’t run the servers, connect a channel or upload anything, so there’s no speed or quality ranking.

ChatCut publishes this guide. ChatCut isn’t a YouTube MCP server, so it appears only in the last section, on editing your own footage before upload, and is described on the same points. For MCP servers that edit video, see the video editing MCP comparison.

Search and public data servers

These read public YouTube data. None of them documents access to your channel’s private data or changes to your channel.

1. ZubeidHendricks YouTube MCP server

ZubeidHendricks/youtube-mcp-server is the most-starred general-purpose server here: 10 read-only tools for video details, search, transcripts, channels and playlists. MIT license; 597 stars, last updated August 8, 2026.

Setup: add npx -y zubeid-youtube-mcp-server to your client’s MCP config with a YouTube Data API key in YOUTUBE_API_KEY. The README shows configs for Claude Desktop and VS Code.

Cost: free code. Requests count against your Google Cloud project’s YouTube Data API quota; you can add backup keys that it switches to when one runs out.

Watch for: it doesn’t document uploads or owner analytics.

2. vidlens-mcp for a research library

thatsrajan/vidlens-mcp lets you import videos once, then search their transcripts, comments, frames and on-screen text (OCR) later. MIT license; 36 stars, last updated October 10, 2026.

Setup: follow the README; it needs no API keys to start.

Cost: free code; some optional features use other services, so check those before turning them on.

Watch for: ask which source each finding came from: speech, a comment, a frame or OCR.

3. Anarcyst server for no-key research

Anarcyst/youtube-mcp-server searches videos, gets transcripts and browses channels without API keys or sign-in. MIT license; 13 stars, last updated March 17, 2026.

Setup: follow the README’s install steps.

Cost: free code; no Google quota involved.

Watch for: the maintainer warns that transcript requests can be rate-limited after many requests in a short time. Test one video before a batch.

4. TranscriptAPI for hosted research

ZeroPointRepo/youtube-mcp connects to TranscriptAPI, a hosted service for transcripts, video search, channel browsing and playlist extraction. MIT license; 13 stars, last updated September 21, 2026.

Setup: sign in with a TranscriptAPI account through OAuth 2.1, or use an API key.

Cost: the README lists a Free plan plus Starter Monthly and Starter Annual credit plans. The MIT code doesn’t make the service free.

Watch for: this is a vendor account, not access to your Google channel. Speed claims in the README are the vendor’s own.

For any server that does search and transcripts, start with a small, source-linked request:

Try this prompt
Find three public YouTube videos about recording clear speech in an untreated room. Return each video's title, channel and URL. Retrieve available transcripts, then identify the advice about microphone distance. Include timestamps only if the tool returns them, mark missing information as unavailable, and link every finding to its source. Do not infer what appears on screen from transcript text.

Transcript servers

A transcript tells you what was said, not what’s on screen. If a video has no captions, check whether the server has a fallback. To get a transcript by hand, see how to get a YouTube transcript.

5. kimtaeyoon83 transcripts with language fallback

kimtaeyoon83/mcp-server-youtube-transcript fetches a transcript in the language you ask for and falls back automatically when it isn’t available. MIT license; 603 stars, last updated July 21, 2026.

Setup: the transcript tool works without any key. An optional analyze_video tool needs a TwelveLabs API key.

Cost: free code; TwelveLabs analysis runs on your TwelveLabs account.

Watch for: note which language you actually got back.

6. anaisbetts mcp-youtube for subtitles

anaisbetts/mcp-youtube downloads YouTube subtitles with yt-dlp on your computer. MIT license; 552 stars, last updated August 16, 2026.

Setup: install yt-dlp, then follow the short README.

Cost: free; runs locally.

Watch for: the README is very brief and doesn’t cover authentication. It only works on videos that have subtitles.

7. jkawamoto transcripts with timestamps

jkawamoto/mcp-youtube-transcript returns raw transcript text, timestamped segments and video metadata in multiple languages. MIT license; 496 stars, last updated November 9, 2026.

Setup: follow the README, which documents no Google sign-in. Optional proxy settings need their own credentials.

Cost: free code; a proxy service, if you use one, is billed separately.

Watch for: the maintainer warns that YouTube blocks automated transcript requests, especially from cloud and datacenter IP addresses. Test from the machine that will run it.

8. TranscriptOut for batch transcripts

artemchuikin/youtube-mcp connects to TranscriptOut, a hosted service with transcripts in five formats, video and channel search, and batch jobs of up to 4,000 transcripts. MIT license; 1 star, last updated October 15, 2026.

Setup: add your TranscriptOut API key as a Bearer token.

Cost: the README lists Free, Starter, Starter Annual and Scale credit plans.

Watch for: run a small batch first and check how it handles videos without transcripts.

9. SeanPedersen transcripts in five languages

SeanPedersen/youtube-transcript-mcp fetches transcripts in English, German, Spanish, French or Russian. 8 stars, last updated May 8, 2026; GitHub detects no license.

Setup: follow the README; it doesn’t state any authentication requirement.

Cost: free to run, but without a license you have no stated permission to reuse the code.

Watch for: it adds an instruction asking the model to strip advertising from the transcript. Keep the raw transcript if you need to check what was removed.

10. Eli-xir with local Whisper fallback

Eli-xir/youtube-transcript-mcp uses captions first and falls back to local speech recognition with faster-whisper when captions are missing. 2 stars, last updated August 16, 2026; GitHub detects no license.

Setup: install the local dependencies in the README. Transcription uses no API keys.

Cost: free; speech recognition uses your computer’s processing power.

Watch for: test one video with captions and one without, and check which route produced the text.

11. youtube-mcp-cli for text output

umbertotancorre/youtube-mcp-cli is an MCP server and command-line tool that returns a transcript as plain text or timestamped segments, all locally. MIT license; 33 stars, last updated August 10, 2026.

Setup: follow the README; it needs no API keys.

Cost: free; runs locally.

Watch for: choose the timestamped format if you plan to quote passages, and check them against the video.

Upload and channel management servers

These change your channel, so they need your Google account’s permission. Review the target channel, title, visibility and schedule before every write.

12. yutu for channel management

eat-pray-ai/yutu is a command-line tool, MCP server and agent for YouTube workflows: uploading videos, managing captions, comments, playlists, thumbnails and channel branding. Apache-2.0 license; 723 stars, last updated November 11, 2026.

Setup: create a Google Cloud project, enable YouTube Data API v3, create a Desktop OAuth client and run yutu auth. Then yutu mcp setup --agents claude,codex adds it to your agents. Install with Homebrew (brew install yutu), WinGet or Go.

Cost: free code; requests use your Google Cloud project’s YouTube Data API quota.

Watch for: its auth code requests the youtube, youtube.force-ssl and youtube.channel-memberships.creator scopes, which let it manage your channel. Keep the token file private.

13. youtube-studio-mcp for owner tools

mateogasme/youtube-studio-mcp handles resumable uploads, changes to video metadata, status and thumbnails, deletion, and channel analytics including demographics, traffic sources and revenue. MIT license; 0 stars, last updated May 25, 2026.

Setup: enable the YouTube Data and Analytics APIs and create a Desktop OAuth client. Its scope code requests youtube, youtube.force-ssl, youtube.upload and yt-analytics.readonly by default, and adds yt-analytics-monetary.readonly when enableMonetary is on.

Cost: free code; requests use your Google API quota.

Watch for: it can delete videos, and it has no GitHub stars yet, so test it on a channel you don’t mind experimenting with.

14. Zernio for hosted publishing

Zernio’s YouTube MCP is a hosted service that uploads and schedules videos and Shorts, and reads captions and analytics for channels you connect. It’s a hosted product; we didn’t find a public source repository.

Setup: connect your YouTube account in Zernio, then sign your agent in with OAuth or an API key.

Cost: Zernio describes the MCP server as free, with billing per connected account.

Watch for: its transcript and analytics tools only cover channels you connect, not other creators’ videos. We didn’t find a published list of the Google scopes it requests; read the consent screen before approving.

Need the Shorts before you upload? Cut them in ChatCut.Download DesktopDownload for macOSDownload for WindowsDownload ChatCut Desktop

Before any upload, have the agent show you exactly what it will send:

Try this prompt
Use the finished local video I select to draft a YouTube upload proposal. Show the target channel, title, description, file and proposed private visibility. List any setting you cannot determine. Do not upload, schedule, publish or change channel data until I approve this exact proposal.

Owner analytics servers

Your channel’s analytics are private, so they need OAuth. A public search result never shows that an agent can read your reports. youtube-studio-mcp, yutu and Zernio above also touch analytics or channel data; youtube-kit is the one built around analytics.

15. youtube-kit for owner analytics

eukosh/youtube-kit is a command-line tool and MCP server for owner analytics, including audience retention, plus resumable uploads. MIT license; 3 stars, last updated November 1, 2026.

Setup: owner features use Google OAuth, with a separate login per channel; public data (search, videos, channels, playlists, comments) works with a YOUTUBE_API_KEY.

Cost: free code; requests use your Google API quota.

Watch for: its auth code requests youtube.force-ssl and yt-analytics-monetary.readonly by default, so it asks for revenue-report access even if you only want retention.

Try this prompt
Using only the connected channel's authorized analytics tools, retrieve an audience-retention report for one video I select. State the channel, video, date range and fields actually returned. Flag missing metrics and authorization errors instead of estimating them. Suggest two passages to review in the original recording, and make no channel changes.

Choose access and Google scopes

Ask three questions before connecting a server:

  1. Does it need a Google API key? Google’s authorization guide uses API keys for requests that only read public data. Every request, even an invalid one, costs at least one point of your project’s quota. No-key transcript servers use their own route instead, and can be rate-limited or blocked.
  2. Does it need your channel’s permission? Uploads, edits and private data need OAuth. Compare the scopes a server requests with what you actually want it to do.
  3. Does it need a vendor account? TranscriptAPI, TranscriptOut and Zernio have their own sign-in. A vendor token doesn’t tell you which Google permissions a connected channel grants.

For analytics, Google’s Reports reference describes yt-analytics.readonly for activity reports and yt-analytics-monetary.readonly for reports that include estimated revenue and ad performance. The same page currently says the Reports query method requires youtube.readonly, which neither youtube-studio-mcp nor youtube-kit lists by default. Run one real report before you build a workflow on either.

Turn your own video into Shorts with ChatCut

The servers above find, read and publish videos. Cutting a long video into Shorts happens in between, in an editor. ChatCut lets Claude Code, Codex and other agents edit videos in ChatCut Desktop: you describe the result in the agent, ChatCut shows the real timeline, and you can adjust it yourself at any time.

The Long Video to Shorts Skill cuts one long podcast, interview, course or livestream into stand-alone shorts. Each is planned to open with a hook, keep enough context to make sense alone and end after its payoff. Give it the video as an upload, a link (its form accepts YouTube links) or a clip already in your project, and choose up to 10 clips, a target length (15 to 90 seconds, or Auto), vertical 9:16 or landscape 16:9, and extras such as captions and title text. It shows a clip plan with source ranges before it cuts. Use videos you own or have permission to edit.

Demo
One long video link becomes a set of captioned vertical shorts — from the Long Video to Shorts Skill.

Setup: install ChatCut Desktop on macOS 13 or later (Apple silicon or Intel) or Windows 10 or later (x64). In a new Claude Code or Codex conversation on that computer, send Read ChatCut.io/desktop to install ChatCut, sign in to ChatCut if it asks, then start a new conversation so the agent loads ChatCut’s tools. Then ask:

Try this promptOpen in ChatCut
Use Long Video to Shorts on the long recording in this ChatCut project. Find three self-contained moments that keep the speaker's meaning, 45 to 60 seconds each, vertical 9:16 with captions. Show me the clip plan with source ranges and hooks before you cut anything.

Cost: the Free plan includes the core editor: manual timeline editing, uploads, transcription and standard caption editing, which never use credits, and exporting doesn’t use credits either. Run from Claude Code or Codex, the agent’s work is covered by your Claude or ChatGPT plan; run from ChatCut’s built-in Agent, its messages use credits. ChatCut-hosted generation such as music or voiceover uses credits too.

Watch for: ChatCut’s docs don’t cover publishing to YouTube. Export the shorts, then upload them in YouTube Studio or with one of the upload servers above.

Review your own recording as short clip timelines in ChatCut.Download DesktopDownload for macOSDownload for WindowsDownload ChatCut Desktop

Questions about YouTube MCP servers

What is a YouTube MCP server?
It's a server that gives an AI agent YouTube tools through the Model Context Protocol, so the agent can search videos, fetch transcripts, upload to your channel or read your analytics while you talk to it. Each server covers some of these jobs, not all.
Is a YouTube MCP server free?
Most self-hosted servers are free, open-source code. Servers that use the YouTube Data API spend your Google Cloud project's quota, and hosted services such as TranscriptAPI, TranscriptOut and Zernio have their own free and paid plans. Your AI agent's plan is a separate cost.
Which YouTube MCP server can upload to my channel?
yutu, youtube-studio-mcp, youtube-kit and Zernio document uploads. All need permission to your Google account; check the scopes and have the agent show the upload settings before it sends anything.
Can a YouTube MCP get another creator's transcript?
The public transcript servers, such as kimtaeyoon83's, jkawamoto's or ZubeidHendricks', fetch transcripts for public videos when captions exist. Rate limits and IP blocking can stop requests. Zernio's transcript tools only cover channels you connect.
How do I connect a YouTube MCP to Claude?
Follow the server's own instructions for your Claude client. For example, the ZubeidHendricks README shows a Claude Desktop config that runs npx -y zubeid-youtube-mcp-server with your API key, and yutu has a yutu mcp setup command that configures Claude and Codex.
Can I use one without self-hosting?
Yes. TranscriptAPI and TranscriptOut are hosted transcript services, and Zernio is a hosted publishing and analytics service. Each needs its own account.
Can I publish Shorts with a YouTube MCP?
Zernio advertises uploading and scheduling Shorts, and the self-hosted upload servers can upload a finished video file. Cut the Shorts first in an editor, then upload.
Is ChatCut free for cutting Shorts?
The Free plan includes the core editor: manual timeline editing, uploads, transcription and standard caption editing, which never use credits. Exporting doesn't use credits; on Free, web and cloud exports share a cumulative 60-minute account quota, while ChatCut Desktop renders video locally. With Claude Code or Codex, the agent's work is covered by your Claude or ChatGPT plan, not ChatCut credits. Messages to ChatCut's built-in Agent and ChatCut-hosted generation use credits.

Checking your footage...

Less editing. More creating.

Download ChatCut Desktop free and make your first video today.

Download Desktop Download for macOS Download for Windows Try it for free