A YouTube MCP server lets your agent fetch current YouTube data instead of relying on training data. It can search for a video, read its transcript, inspect a channel, or retrieve comments through the same tool interface it already uses.
We reviewed the current documentation, source repositories, install paths, and pricing for seven options. We did not run a synthetic latency benchmark, so this comparison does not make performance claims it cannot support.
Updated July 2026. Features and prices change. Follow the linked source for each server before you commit to one.
The short answer
- Pick Stophy when your agent needs broad YouTube coverage through a hosted MCP server: search, transcripts, comments, replies, channels, playlists, autocomplete, live chat, YouTube Music, and YouTube Kids.
- Pick mcp-youtube by kirbah when you want token-efficient Data API tools and can provide a Google API key for features beyond transcripts.
- Pick mrsknetwork/ytmcp when you need public Data API tools plus optional OAuth access to your own subscriptions or memberships.
- Pick Firecrawl when it is already your general web-data layer and YouTube transcripts are a secondary use case.
- Pick @inlustris/youtube-mcp when you want a local, no-key server with transcripts, search, channel discovery, and video frames.
| Server | Transport | Authentication | Main coverage | Best fit |
|---|---|---|---|---|
| Stophy | Hosted HTTP or local stdio | Stophy API key | Eight tools for search, video data, channels, playlists, suggestions, Music, Kids, and credits | Production agents that need broad YouTube coverage |
| mcp-youtube | stdio | None for transcripts; Google API key for other data | Transcripts, search, videos, channels, trends, comments | Token-efficient Data API workflows |
| mrsknetwork/ytmcp | stdio | None for transcripts; API key or OAuth for more | Transcripts, search, metadata, comments, playlists, account data | Teams that need public and owner-authorized data |
| Firecrawl MCP | Hosted HTTP or local stdio | Firecrawl API key or OAuth token | General web search and scraping, including YouTube transcripts | Existing Firecrawl users |
| iamkw0n/youtube-mcp | Local stdio | None; optional Zyte key | Search, videos, channels, playlists, comments | Local page-scraping workflows |
| @inlustris/youtube-mcp | Local stdio | None | Transcripts, video frames, video and channel search | Transcript and visual grounding |
| youtube-transcript-mcp | Hosted SSE or HTTP | None | One transcript tool with language selection | Zero-setup transcript retrieval |
How to compare YouTube MCP servers
Start with the data your agent must retrieve.
- Transcript output. Check for timestamps, language selection, and a clear response when captions do not exist.
- Coverage. Decide whether you also need search, channel data, playlists, comments, live chat, Music, or Kids.
- Transport. Hosted HTTP works in environments that cannot keep a local process running. Stdio is simple when the MCP client runs on your machine.
- Authentication. A no-key server reduces setup. An API-backed server takes more setup but can centralize reliability, billing, and support.
- Operating limits. Check Google quota, vendor credits, request limits, and the failure modes of unofficial page parsing.
This review uses first-party repositories and documentation. It removes marketplace ratings and download counts because those numbers change quickly and do not prove that a server fits your workload.
1. Stophy
Best for: Agents that need a hosted YouTube data layer with more than transcripts.
Stophy exposes eight MCP tools:
stophy_search_videossearches videos, Shorts, channels, and playlists with filters.stophy_get_videoretrieves details, transcripts, comments, replies, or live chat.stophy_get_channelretrieves videos, Shorts, live streams, playlists, posts, or channel details.stophy_get_playlistretrieves playlist metadata and videos.stophy_get_suggestionsreturns YouTube autocomplete suggestions.stophy_musicsearches YouTube Music and retrieves songs, lyrics, albums, artists, and playlists.stophy_kidssearches YouTube Kids and retrieves Kids video data.stophy_get_creditsreturns the remaining balance without consuming a credit.
The video tool returns timestamped transcript segments and supports comments, replies, and live-chat continuation tokens. Stophy is the only server in this reviewed set with dedicated MCP tools for YouTube Music and YouTube Kids. It is also the only one in the set with live chat as a first-class video operation.
That distinction needs one qualification: Google does expose live-chat resources through the YouTube Live Streaming API. The comparison here is about the tools each MCP server currently exposes, not a claim that Google's APIs have no live-chat support.
Connect over hosted HTTP:
$ claude mcp add --transport http stophy https://api.stophy.dev/v1/mcp \
--header "Authorization: Bearer $STOPHY_API_KEY"Or run the local stdio package:
$ claude mcp add stophy -e STOPHY_API_KEY=st_YOUR_API_KEY -- npx -y @stophy/mcpPricing starts with 100 free credits and no card. Paid packs are $9 for 3,000 credits, $25 for 10,000, $70 for 50,000, and $200 for 200,000. Credits do not expire. Failed requests are refunded.
Tradeoffs: Stophy only covers YouTube. It relies on captions supplied through YouTube, so it does not transcribe the audio of a video that has no captions.
2. mcp-youtube by kirbah
Best for: Agents that need compact Data API responses and transcript access through a local server.
mcp-youtube runs without credentials when you only need transcripts. Add a YouTube Data API v3 key to unlock search, video details, channel statistics, trends, categories, and comments. MongoDB is optional and enables caching plus its outlier-channel analysis tool.
$ npx -y @kirbah/mcp-youtubeThe server filters responses for language-model use instead of forwarding every field from Google. That can reduce context use when an agent makes several calls.
Tradeoffs: Most features require your own Google Cloud project and API key. Data API calls consume Google quota, while the transcript tool uses a separate no-key path. It does not expose dedicated Music, Kids, or live-chat tools.
3. mrsknetwork/ytmcp
Best for: Agents that need both public YouTube data and optional account-authorized tools.
mrsknetwork/ytmcp has three access levels:
- Guest mode uses
yt-dlpfor transcripts and needs no credentials. - An API key unlocks public search, metadata, channel, playlist, comment, caption-list, category, and region tools.
- OAuth unlocks data tied to the signed-in account, including subscriptions and channel memberships where available.
$ npx -y @mrsknetwork/ytmcp@latestTradeoffs: Search and most structured data tools consume your Google project quota. OAuth adds a browser authorization flow and is only necessary for account-specific data. There are no dedicated Music, Kids, or live-chat tools.
4. Firecrawl MCP
Best for: Agents that already use Firecrawl for web search and scraping.
Firecrawl MCP is a general web-data server. Its scrape tool can return a YouTube transcript when captions are available, but it does not model YouTube channels, comments, replies, Music, Kids, or live chat as dedicated tools.
Run it locally:
$ FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcpOr use Firecrawl's documented hosted MCP endpoint. The Firecrawl CLI command sometimes shown in setup guides installs Firecrawl's agent skill; it is not the MCP server installation command.
The current Firecrawl pricing includes 1,000 credits per month on the free plan. Paid plans start at $16 per month when billed yearly. Standard plan credits do not roll over.
Tradeoffs: YouTube is one input among many. Choose Firecrawl for a shared web layer, not for a YouTube-specific schema.
5. iamkw0n/youtube-mcp
Best for: Local agents that need public YouTube pages without a Google API key.
iamkw0n/youtube-mcp parses public YouTube page data. It exposes tools for video search, video details, channels, playlists, and comments. It does not currently list a transcript tool.
The repository is explicit about its limits: changes to YouTube's page structure can break the parser, and excessive requests can be blocked. It supports an optional Zyte proxy for that case.
Tradeoffs: You must clone and build it before pointing your MCP client at the local server. The repository recommends personal, non-commercial use. Treat it as a lightweight local option, not a drop-in hosted dependency.
6. @inlustris/youtube-mcp
Best for: Local transcript retrieval with optional video-frame capture.
The project in adshaa/youtube-mcp now installs as @inlustris/youtube-mcp. Its five tools retrieve timestamped transcripts, capture a video frame, search videos, search channels, and list channel videos. Transcript output can be plain text, and the tool can remove SponsorBlock segments when requested.
$ npx -y @inlustris/youtube-mcp@latestFull-resolution frame capture needs yt-dlp and FFmpeg. Without them, the tool falls back to YouTube storyboard images.
Tradeoffs: It does not expose comments, playlists, live chat, Music, or Kids. The current project does not advertise built-in transcript chunking, so chunk the timestamped output in your own RAG pipeline.
7. youtube-transcript-mcp by ergut
Best for: A hosted transcript tool with no account or local server setup.
youtube-transcript-mcp runs on Cloudflare Workers and exposes one tool: get_transcript. It accepts a YouTube URL and an optional language code. The public server supports SSE and HTTP without authentication.
Tradeoffs: It only retrieves transcripts. It has no search, metadata, comments, channel, playlist, live-chat, Music, or Kids tools. The public endpoint is maintained by a community project, so confirm its availability and data-handling requirements before putting it in a production workflow.
Pricing and operating model
| Server | Starting cost | What limits usage |
|---|---|---|
| Stophy | 100 free credits; $9 for 3,000 | Purchased credits; they do not expire |
| mcp-youtube | Open-source server | Google quota for Data API tools; your compute |
| mrsknetwork/ytmcp | Open-source server | Google quota for API tools; your compute |
| Firecrawl MCP | 1,000 free monthly credits; paid from $16/month yearly | Monthly credits and plan limits |
| iamkw0n/youtube-mcp | Open source | Your compute and YouTube page-scraping limits |
| @inlustris/youtube-mcp | Open source | Your compute and upstream availability |
| youtube-transcript-mcp | Public hosted endpoint | Maintainer availability and upstream limits |
Google's quota rules are method-specific and can change. Check the YouTube Data API quota calculator instead of treating every MCP call as equal.
Which one should you choose?
Choose against your narrowest real requirement:
- Need hosted search, transcripts, comments, channels, playlists, live chat, Music, or Kids in one connection: Stophy.
- Need local, token-efficient Data API tools: mcp-youtube.
- Need your own subscriptions or channel-membership data: mrsknetwork/ytmcp with OAuth.
- Need YouTube transcripts inside an existing general web-scraping stack: Firecrawl.
- Need local search and public page data without a key: iamkw0n/youtube-mcp.
- Need transcripts plus video frames without a key: @inlustris/youtube-mcp.
- Need one hosted transcript tool with no account: youtube-transcript-mcp.
No-key does not mean no operational risk. Servers that use public pages, InnerTube, or yt-dlp depend on interfaces that YouTube can change or restrict. Data API wrappers trade that risk for project setup and quota. Hosted vendors trade local maintenance for credits and vendor dependence.
Frequently asked questions
Can a YouTube MCP server return a transcript for every video?
No. Caption-based tools need an accessible caption track. A private, deleted, restricted, or uncaptioned video may return no transcript. None of the seven servers reviewed here promises audio transcription as a fallback for every captionless video.
Does the official YouTube Data API return third-party transcripts?
The official captions endpoints require OAuth authorization. Listing caption tracks and downloading the caption file are separate operations, and access is constrained by the video's authorization context. Google's captions implementation guide documents those requirements. Several MCP servers therefore use public transcript paths instead of the captions API.
Which server supports live chat?
Stophy is the only server in this reviewed set that lists live chat as a first-class MCP operation. Google's separate Live Streaming API also exposes live-chat messages, so you can build or choose another wrapper around that API if it fits your authorization and quota requirements.
Which server supports YouTube Music or YouTube Kids?
Stophy is the only server in this set with dedicated MCP tools for both surfaces. The YouTube Data API v3 does not provide dedicated Music or Kids catalog endpoints.
Do these servers work with Claude Code?
They use MCP, but the transport determines the setup. Claude Code can connect directly to hosted HTTP servers such as Stophy. Local stdio servers run as child processes through commands such as npx. Hosted SSE servers may need a bridge such as mcp-remote, depending on the client version.
Is a paid server better than a free server?
It depends on the workload. A local open-source server can be enough for occasional research. A hosted server can remove deployment, proxy, caching, and upstream maintenance work. Compare the total operating burden, not only the first request price.
Start with Stophy
Create an account with 100 free credits, connect the hosted MCP endpoint, and test it against the videos and channels your agent will actually process.
Related guides: YouTube transcript API comparison, YouTube Data API alternatives, and YouTube transcripts in Python.
Bias disclosure: This comparison was written by the team behind Stophy. Stophy appears first because the comparison prioritizes the breadth and hosted transport we built. Every competitor description links to its current first-party repository or documentation so you can check the claims directly.