YouTube MCP Server
Connect your AI agents to HasData's YouTube MCP server. Scrape channel stats, video metadata, and timed transcripts in structured JSON over streamable HTTP.
https://mcp.hasdata.com/api/mcp?apis=youtube - Transport
- Streamable HTTP
- Auth
- OAuth or API key
- Tools
- 4 YouTube tools
- Price
- 10 credits / call
Data teams at
Playground
See the YouTube MCP server in action
Select a sample prompt to inspect the tool it chooses and the JSON payload your agent receives.
YouTube · 4 tools Enter to send
Find the most-viewed videos about the model context protocol from the last month
113,347 results. Three most viewed of the last month:
MCP Was Wrong From The Start (They Just Fixed It) · 170,065 views · 6:52
Claude Code Full Course · 161,797 views · 1:21:37
Did Anthropic finally fix MCP? · 102,650 views · 18:17
Views come back twice, as the text YouTube shows and as an integer you can sort on.
youtube_search { "q": "model context protocol", "sortBy": "views", "date": "month" }
{
"url": "https://api.hasdata.com/scrape/youtube/search?q=model+context+protocol&sortBy=views",
"status": 200,
"json": {
"searchInformation": { "totalResults": 113347 },
"videoResults": [
{ "positionOnPage": 0, "videoId": "f4mI3d-nTrI",
"title": "MCP Was Wrong From The Start (They Just Fixed It)",
"viewsOriginal": "170K views", "views": 170065, "length": "6:52",
"publishedDate": "3 weeks ago" }
],
"pagination": { "nextPageToken": "eyJjb250aW51YXRpb24iOiJFcW9ERWhadGIy…" }
}
}
Find the most-viewed videos about the model context protocol from the last month
113,347 results. Three most viewed of the last month:
MCP Was Wrong From The Start (They Just Fixed It) · 170,065 views · 6:52
Claude Code Full Course · 161,797 views · 1:21:37
Did Anthropic finally fix MCP? · 102,650 views · 18:17
Views come back twice, as the text YouTube shows and as an integer you can sort on.
youtube_search { "q": "model context protocol", "sortBy": "views", "date": "month" }
{
"url": "https://api.hasdata.com/scrape/youtube/search?q=model+context+protocol&sortBy=views",
"status": 200,
"json": {
"searchInformation": { "totalResults": 113347 },
"videoResults": [
{ "positionOnPage": 0, "videoId": "f4mI3d-nTrI",
"title": "MCP Was Wrong From The Start (They Just Fixed It)",
"viewsOriginal": "170K views", "views": 170065, "length": "6:52",
"publishedDate": "3 weeks ago" }
],
"pagination": { "nextPageToken": "eyJjb250aW51YXRpb24iOiJFcW9ERWhadGIy…" }
}
}
Give me the stats on the video "MCP Was Wrong From The Start"
170,126 views, 3,813 likes, 6 minutes 52 seconds.
Better Stack, 197K subscribers. Published 8 August 2026 under Science & Technology.
Twenty-eight tags, starting with mcp and model context protocol.
It also returns an aiSummary of the video and the caption tracks available, here one auto-generated English track.
youtube_video { "v": "f4mI3d-nTrI" }
{
"url": "https://api.hasdata.com/scrape/youtube/video?v=f4mI3d-nTrI",
"status": 200,
"json": {
"videoId": "f4mI3d-nTrI", "title": "MCP Was Wrong From The Start (They Just Fixed It)",
"views": "170,126 views", "extractedViews": 170126,
"likes": "3.8K", "extractedLikes": 3813,
"lengthSeconds": 412, "publishedDate": "Aug 8, 2026",
"category": "Science & Technology",
"channel": { "name": "Better Stack", "extractedSubscribers": 197000 },
"captions": [{ "languageCode": "en", "name": "English (auto-generated)", "kind": "asr" }],
"aiSummary": "Better Stack explores the shift to a stateless Model Context Protocol…"
}
}
Summarise what "MCP Was Wrong From The Start" actually says
227 timed cues, and eleven chapters to hang them on.
Intro runs 0:00 to 0:22, then "Why Stateful MCP Broke" to 1:21.
It opens: "MCP just shipped its biggest change since it launched, and the headline is…"
Every cue has a start and an end in milliseconds plus the timestamp as displayed.
youtube_transcript { "v": "f4mI3d-nTrI", "languageCode": "en" }
{
"url": "https://api.hasdata.com/scrape/youtube/transcript?v=f4mI3d-nTrI&languageCode=en",
"status": 200,
"json": {
"transcript": [
{ "startMs": 0, "endMs": 1600, "startTimeText": "0:00",
"snippet": "MCP just shipped its biggest change" }
],
"chapters": [
{ "chapter": "Intro", "startMs": 0, "endMs": 22000 },
{ "chapter": "Why Stateful MCP Broke", "startMs": 22000, "endMs": 81000 }
],
"availableTranscripts": [{ "languageCode": "en", "type": "asr", "selected": true }]
}
}
What else has the Better Stack channel posted?
Thirty uploads on the first page of the videos tab.
Fable 5.1 is here, and its REALLY good
This 2B Model Cloned My Voice Locally (VoxCPM)
Three tokens come back with the page: the next one, and separate ones for the popular and the oldest orderings.
youtube_channel { "channelId": "UCkVfrGwV-iG9bSsgCbrNPxQ", "tab": "videos" }
{
"url": "https://api.hasdata.com/scrape/youtube/channel?channelId=UCkVfrGwV-iG9bSsgCbrNPxQ&tab=videos",
"status": 200,
"json": {
"channelInfo": { "channelId": "UCkVfrGwV-iG9bSsgCbrNPxQ", "name": "Better Stack",
"handle": "@betterstack", "rssUrl": "https://…", "availableTabs": ["videos"] },
"videos": [{ "videoId": "0lBvjhcRqyU", "title": "Fable 5.1 is here, and its REALLY good" }],
"pagination": { "nextPageToken": "eyJ…", "popularPageToken": "eyJ…", "oldestPageToken": "eyJ…" }
}
}
Tools
The YouTube tools your agent can call
4 toolsTyped parameters, required flags, and structured JSON fields for each tool. The transcript tool is the one with no equivalent in the official API, and it returns the timed track rather than a wall of text.
+ hasdata_youtube_search_getYoutubeSearchResults Search with the real filter set 5 params 10 credits
Try| Parameter | Type | Required | Example |
|---|---|---|---|
| q | string | required | "model context protocol" |
| sortBy | enum | optional | "views" |
| date | enum | optional | "month" |
| length | enum | optional | "plus20" |
| videoType | enum | optional | "video" |
response
- searchInformation
- videoResults
- videoId
- title
- views
- length
- publishedDate
- pagination
+ hasdata_youtube_video_getYoutubeVideo One video's metadata by id 3 params 10 credits
Try| Parameter | Type | Required | Example |
|---|---|---|---|
| v | string | required | "dQw4w9WgXcQ" |
| gl | string | optional | "US" |
| hl | string | optional | "en" |
response
- title
- extractedViews
- extractedLikes
- lengthSeconds
- publishedDate
- category
- keywords
- channel
- captions
- aiSummary
+ hasdata_youtube_transcript_getYoutubeTranscript The timed transcript of a video 3 params 10 credits
Try| Parameter | Type | Required | Example |
|---|---|---|---|
| v | string | required | "dQw4w9WgXcQ" |
| languageCode | string | optional | "en" |
| type | enum | optional | "asr" |
response
- transcript
- startMs
- endMs
- startTimeText
- snippet
- chapters
- availableTranscripts
+ hasdata_youtube_channel_getYoutubeChannel Channel identity and one tab 3 params 10 credits
Try| Parameter | Type | Required | Example |
|---|---|---|---|
| channelId | string | required | "UCLA_DiR1FfKNvjuUpBHmylQ" |
| tab | enum | optional | "videos" |
| paginationToken | string | optional | "4qmFsgJ…" |
response
- channelInfo
- name
- handle
- description
- rssUrl
- availableTabs
- videos
- pagination
Every call is 10 credits. Failed calls are not billed.
Connect
One endpoint, any MCP client
Config blocks for Claude Desktop, Cursor, VS Code, Windsurf and Claude Code, in each one’s native format. Authenticate with browser OAuth or an x-api-key header.
Claude Code
$claude mcp add --transport http hasdata-youtube \
> https://mcp.hasdata.com/api/mcp?apis=youtube \
> --header "x-api-key: YOUR_API_KEY"
· One line, no restart. Run /mcp to confirm 4 tools.
Check the tools are thereIn the session
/mcp
hasdata-youtube · connected · 4 toolsCursor
{
"mcpServers": { "hasdata-youtube": {
"url": "https://mcp.hasdata.com/api/mcp?apis=youtube",
"headers": { "x-api-key": "YOUR_API_KEY" }
} }
}
· Add this entry to ~/.cursor/mcp.json. Servers already in the file stay as they are.
· Project-level config also works at .cursor/mcp.json
Check the tools are thereAfter saving the file
Settings → MCP
hasdata-youtube · 4 tools enabledClaude Desktop
Recommended · OAuth connector
1.Settings → Connectors → Add custom connector
2.Paste https://mcp.hasdata.com/api/mcp?apis=youtube
3.Sign in with HasData
{
"mcpServers": { "hasdata-youtube": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.hasdata.com/api/mcp?apis=youtube",
"--header", "x-api-key:YOUR_API_KEY"]
} }
}
· Add this entry to ~/Library/Application Support/Claude/claude_desktop_config.json. Servers already in the file stay as they are.
· The config file only takes stdio servers, so a static key rides the mcp-remote bridge. On Windows the file lives at %APPDATA%\Claude\claude_desktop_config.json.
Check the tools are thereReopen the app
Settings → Connectors
hasdata-youtube · connected · 4 toolsVS Code / Copilot
{
"servers": { "hasdata-youtube": {
"type": "http",
"url": "https://mcp.hasdata.com/api/mcp?apis=youtube",
"headers": { "x-api-key": "YOUR_API_KEY" }
} }
}
· Add this entry to .vscode/mcp.json. Servers already in the file stay as they are.
· Agent mode picks the server up on save.
Check the tools are thereAgent mode
Ctrl/⌘ + Shift + P → MCP: List Servers
hasdata-youtube · running · 4 toolsWindsurf
{
"mcpServers": { "hasdata-youtube": {
"serverUrl": "https://mcp.hasdata.com/api/mcp?apis=youtube",
"headers": { "x-api-key": "YOUR_API_KEY" }
} }
}
· Add this entry to ~/.codeium/windsurf/mcp_config.json. Servers already in the file stay as they are.
· Windsurf reads serverUrl, not url. Header auth only.
Check the tools are thereIn Cascade
Cascade → MCP servers
hasdata-youtube · 4 toolsCodex CLI
[mcp_servers.hasdata-youtube]
url = "https://mcp.hasdata.com/api/mcp?apis=youtube"
env_http_headers = { "x-api-key" = "HASDATA_API_KEY" }
· Add this entry to ~/.codex/config.toml. Servers already in the file stay as they are.
· TOML, not JSON. Codex reads the key from the HASDATA_API_KEY environment variable, so export it in the shell that starts Codex.
Check the tools are thereIn the terminal
codex mcp list
hasdata-youtube · 4 toolsRaw HTTP / curl
$curl -N https://mcp.hasdata.com/api/mcp?apis=youtube \
> -H "x-api-key: YOUR_API_KEY" \
> -H "Content-Type: application/json" \
> -H "Accept: application/json, text/event-stream" \
> -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
· Fallback for any client not listed. Streamable HTTP wants both Accept types, and leaving one out returns 406.
Check the tools are thereCall it yourself
curl … tools/list
{ "tools": [ … 4 items ] }Scope
YouTube plus whatever else you need
Name the services in the query string and the server exposes only their tools. Every one draws from the same key and the same balance.
https://mcp.hasdata.com/api/mcp?apis=youtube,google_serp,google_trends
15tools in your server
Build vs Buy
Free on GitHub, paid in hours
Anyone can clone a free MCP to scrape YouTube. Whether that stays free depends on the shape of your project.
Scenario
A script I run once
Self-hosted from GitHubyour machine, your upkeep
$0/ mo, runs locally
+ 3–6 h onceof someone’s time
Setup (clone, wire, run)3–6 h once
Markup changesnot yet
Retries and concurrencynot needed
Proxy and hosting$0
Nothing to own. You get the answer and delete the folder.
mcp.hasdata.comFree tier
$0/ mo
+ 0 hof someone’s time
Setup one line
Parser upkeep included
Retries and concurrency included
~25 calls 250 of 1,000 credits
The free tier renews every month. No card.
leans self-hosted
Take the free repo
Nothing has to keep working tomorrow, so upkeep never appears. Both cost nothing, so take whichever is already in front of you.
A weekly report
Self-hosted from GitHubyour machine, your upkeep
$0–40/ mo
+ 2–4 h / moof someone’s time
Setup (proxy, hosting, wiring)3–6 h once
Markup changes2–4 h / mo
Retries and concurrency2–4 h once
Proxy and hosting$0–40 / mo
Someone looks at it when the numbers come back wrong, which is how you find out.
mcp.hasdata.comFree tier
$0/ mo
+ 0 hof someone’s time
Setup one line
Parser upkeep included
Retries and concurrency included
~100 calls a month 1,000 of 1,000 credits
Exactly the free tier, so a heavier month moves you to Startup.
dead even
Either works
The repo is free if you have somewhere to run it, ours is free up to the credit ceiling. The two to four hours a month decide it.
A pipeline in production
Self-hosted from GitHubyour machine, your upkeep
$40–120/ mo
+ 6–10 h / moof someone’s time
Setup (proxy, hosting, wiring)3–6 h once
Markup changes2–4 h / mo
Retries, concurrency, rate limits4–8 h once
Proxy and hosting$40–120 / mo
Someone owns this permanently, and it is never the top of their list.
mcp.hasdata.comStartup, 200K credits, 5 concurrent
$49/ mo
+ 0 hof someone’s time
Setup one line
Parser upkeep included
Retries and concurrency included
5,000 calls 50K of 200K credits
A quarter of the plan used, and the rest is headroom.
leans hosted
Use the hosted server
The proxy and hosting bill alone is in the range of the whole plan, and that is before anyone spends the six to ten hours.
A feature in my product
Self-hosted from GitHubyour machine, your upkeep
$300+/ mo
+ on-callof someone’s time
Setup (proxy, hosting, wiring)3–6 h once
Markup changes, plus on-call2–4 h / mo
Retries, concurrency, rate limitsongoing
Proxy and hosting$300+ / mo
The thing your customers touch depends on one volunteer's spare evening.
mcp.hasdata.comBasic, 1M credits, 15 concurrent
$99/ mo
+ 0 hof someone’s time
Setup one line
Parser upkeep included
Retries and concurrency included
50,000 calls 500K of 1M credits
Same endpoint at ten calls and at ten million. One invoice.
leans hosted
Use the hosted server
A third of the price, and nobody has to carry a pager for a dependency your customers touch.
Pricing
One balance, every interface
A flat monthly price against a known credit ceiling. Move up a plan when you outgrow it.
Free
$0 /mo
Free forever
100 tool calls / month
1,000 credits / month
1 concurrent request
Every MCP tool on every plan
Failed calls are not billed
One balance across MCP and REST
Community support
Startup
$49 /mo
$2.46 / 1k tool calls
20K tool calls / month
200K credits / month
5 concurrent requests
Every MCP tool on every plan
Failed calls are not billed
One balance across MCP and REST
Email support
Basic
Recommended$99 /mo
$0.99 / 1k tool calls
100K tool calls / month
1M credits / month
15 concurrent requests
Every MCP tool on every plan
Failed calls are not billed
One balance across MCP and REST
Priority email support
Growth
$208 /mo
$0.69 / 1k tool calls
3M credits / month
50 concurrent requests
Every MCP tool on every plan
Failed calls are not billed
One balance across MCP and REST
Dedicated account manager
Monthly tool call volume
Free 100K 500K 2M
Best fit
Basic
tool calls / mo
100K
Concurrency
15
$ / 1k tool calls
$0.99
$99 /mo
Get StartedEnterprise
Custom price based on required volume
Past 20M credits a month, or terms the self-serve plans do not cover. We shape the contract around your workload.
Credits rollover
Unused credits carry into the next billing period.
Concurrency 2000+
Parallel request limits set to your peak load.
#1 request priority
Highest speed, always first in the queue.
Personal manager
A direct line to the founding team.
SSO
SAML single sign-on for the whole team.
SOC-2 compliance
Security review, DPA, and audit reports.
Talk to sales Quote within one business day
The same data, without the agent.
FAQ
Before you wire it in
Do I need a Google account or an API key from YouTube?
No. The only credential is your HasData key. Every tool reads the public pages a signed-out visitor sees.
How is this different from the YouTube Data API?
Two things. The official API runs on a daily quota that a search burns through quickly, and it does not return transcripts at all. These tools have no quota beyond your credit balance, and the transcript is a first-class call.
Which videos have a transcript?
Any video with a subtitle track, and most large channels have one. <code>languageCode</code> picks the track, and <code>type=asr</code> asks for the auto-generated speech-recognition version when a human-written one is missing.
Is the transcript one block of text?
No. Each cue carries its own <code>start</code> and <code>duration</code>, so a summary can point at a timestamp instead of paraphrasing the whole thing.
Can I use the real search filters?
Yes. Upload date, duration, sort order and video type are enums, and <code>filters__</code> covers the rest, including HD, subtitles and live. An agent picks documented values rather than assembling a search URL.
How fresh is the data? Is anything cached?
Every tool call fetches the page at that moment, so view and like counts are the ones on screen right now. Nothing is served from a cache.
How many calls can run in parallel?
One on the free tier, five on Startup, fifteen on Basic, and fifty on Growth. The larger Growth volumes raise it to 200, 300 and 500.
What happens when I run out of credits?
Nothing is charged on top of the plan. Calls stop until the billing cycle renews.
Does it work with Claude Desktop, Cursor, VS Code and Claude Code?
Yes, all four, plus anything else that speaks MCP over streamable HTTP. Claude Desktop connects through OAuth, and the rest take the URL and an API key header.
Is HasData affiliated with YouTube?
No. HasData is an independent service and is not affiliated with, endorsed by, or sponsored by Google LLC. YouTube is a trademark of its respective owner.