YouTube MCP server
YouTube sponsorship data, inside your AI
Ask a model which brands are sponsoring YouTubers right now and it answers from memory. Connect this server and it answers from 250k+ indexed channels and every sponsorship we detected, with a link on every number.
Included on every plan with API access. Lookups are free. Only AI search spends credits.
Remote MCP URL
https://sponsorship.so/mcp
“Who already sponsors @mkbhd, and how often?”
-
dbrand 111
-
Ridge 18
-
ExpressVPN 10
-
Anker 8
-
Eight Sleep 7
218 sponsorships across 1,840 videos. Every row comes back with a link your AI can cite.
Works in the clients you already have open
Real responses
Three things you can ask the YouTube MCP server
Pick one and see the data that came back. These are real responses, not mockups.
“Is NordVPN still buying, and how big is their creator roster?”
Runs
get_brand
{ "brand": "NordVPN" }
8,240
channels sponsored
38,169
sponsored videos
47,759
detected mentions
- Public partner program at nordvpn.com/influencers
- Contact email on file, so your AI can go straight to a pitch
- Same brand profile you'd open in the app, minus the clicking
“What should a mid-roll integration on @Fireship cost?”
Runs
estimate_sponsorship_price
{ "channel": "@Fireship", "sponsorship_type": "integration" }
$36.7k–$56.4k
fair price range
$50–$76
CPM range
582k–895k
projected views
- Opportunity rating 56/100, built from this channel's own view curve
- 3.14% engagement and a 42.86% sponsor renewal rate
- 49 brands have already paid for a slot here
“Which brands sponsor small AI creators?”
Runs
search
{ "query": "brands that sponsor small ai creators", "search_type": "brand" }
256
matching brands
Squarespace
top match, 2,696 channels sponsored
$0
to re-run it later
- A model turns the sentence into filters — niche, size, country, sponsor history
- The search is kept on the workspace, so tomorrow's re-run costs nothing
- Comes back with a web URL that opens the same filters in the app
Trimmed from real responses in August 2026. Your workspace sees today's numbers.
Guessing vs. knowing
The same question, with and without the data
A model sounds just as confident either way. Only one column gives you a number you can open and check.
You ask
Any model on its own
With Sponsorship.so
“Who sponsors this channel?”
Any model on its own
Brands it half-remembers from training data.
With Sponsorship.so
Every sponsorship we detected, with a count of brand mentions.
“What should this cost?”
Any model on its own
A range lifted from someone's 2021 blog post.
With Sponsorship.so
CPM and CPV modeled on that channel's own views.
“Is this brand still active?”
Any model on its own
Nothing after the training cutoff.
With Sponsorship.so
Sponsored videos published in the last 30 days.
“Who else is buying in my niche?”
Any model on its own
A deep research run that browses a few pages and stops.
With Sponsorship.so
Every match across 250k+ indexed channels, in seconds.
“Where did that come from?”
Any model on its own
Nothing to click.
With Sponsorship.so
Every record ships a URL you can open and check.
Setup
Paste one URL. That's the integration.
We host the YouTube MCP server and sign you in over OAuth. When we ship a new tool, your assistant picks it up on the next connection with nothing to redeploy.
1
Paste one URL
There is no package to install and no SDK to keep current. When we ship a tool, your assistant picks it up on the next connection.
https://sponsorship.so/mcp
2
Approve a workspace
Your client sends you to Sponsorship.so to sign in and choose which workspace it may read. Revoke it later from MCP settings.
3
Ask in plain language
No tool names to memorize. Describe a channel, a brand, or a budget and your assistant picks the right call.
Step-by-step instructions for Claude, Cursor, Claude Code, Codex, VS Code, and ChatGPT live in your MCP settings once you're signed in.
Prompts
Questions you can steal
Each one chains several tools together. The manual version of these questions costs an afternoon of tabs and a spreadsheet.
“Price a 60-second mid-roll on @Fireship and tell me whether that CPM is fair for a dev-tools brand.”
“Which brands sponsored my three closest competitors this quarter but have never sponsored me?”
“Shortlist 20 German finance creators under 200k subscribers who already take fintech money, then find emails for the best five.”
“Has dbrand slowed down on YouTube, or have they just moved to smaller channels?”
“Import these 30 handles from my spreadsheet and flag which ones already have sponsors.”
“Pick the three sponsors most likely to say yes to my channel, then draft each pitch around what they already buy.”
The toolbox
19 tools, and what each one costs you
Grouped by the job you're doing, not by our database tables. Most of them are free to call.
AI search
5 credits per new search, re-runs freeDescribe the list you want in plain language and a model builds the filters for you. Every search stays on the workspace, so re-running it later is free.
search
list_searches
get_search
get_credits
Learn about a sponsor
FreeHow much a brand spends on YouTube, who it pays, and which brands behave like it.
get_brand
get_brand_stats
get_similar_brands
get_brand_creators
get_brand_videos
list_most_active_brands
Learn about a creator
Free · 1 credit per new emailAudience numbers, every sponsor they have worked with, lookalike channels, rates, and contact details.
get_youtube_channel
get_channel_sponsors
get_channel_sponsored_videos
get_similar_channels
list_top_creators
find_creator_email
estimate_sponsorship_price
Feed the database
FreePoint us at channels we have not indexed yet and refresh their recent videos.
add_youtube_channels
fetch_youtube_channel_videos
For developers
No MCP client? It's still just HTTP.
Skip OAuth, send an API token, and call the same endpoint over JSON-RPC from n8n, Make, or a cron job. If your code is doing the deciding, the REST API is usually the simpler fit.
curl -X POST https://sponsorship.so/mcp \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_TOKEN" \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "list_most_active_brands",
"arguments": { "tag": "VPN", "since_days": 30 }
}
}'
"Being able to discover creators in bulk, organize campaigns, and estimate charges upfront saves us countless hours every week. It's become an essential tool for our marketing team."
![]()
Nuno Sancha
Marketer at
TRMNL
FAQ
Frequently asked questions
About a minute. Paste the server URL into your assistant, sign in to your workspace, and ask about a channel or brand. The answer comes back in the same chat, with no export and no code.
Who sponsors a channel and how many videos each brand ran, which creators a brand pays, recent sponsored videos, similar channels and brands, price and CPM estimates, rankings, and contact emails. Every record includes a url that opens the same page in the app.
No. Setup is one URL and a sign-in, and you ask in plain language. If you'd rather script it, the same endpoint also accepts a bearer token from n8n, Make, or your own code.
However you already have it written down. Channels accept a handle like @mkbhd, a YouTube ID starting with UC, or our UUID. Brands accept a name or a UUID. If a channel is not indexed yet, asking for it imports it.
Looking things up is free: brand profiles, sponsor history, similar channels, price estimates, rankings, and bulk import. A new AI search costs 5 credits, and a newly discovered email costs 1. Re-running a search you already ran is always free.
It can run searches, save them to your history, add channels for indexing, and spend credits on those two actions. It cannot touch billing, team members, passwords, or anything else in your settings.
Any client that speaks remote HTTP MCP. We use it daily in Claude, Cursor, Claude Code, ChatGPT developer mode, VS Code, and Codex CLI. Stdio-only apps can bridge through mcp-remote.
Get sponsorship data in your assistant
Connect the server once, then ask who sponsors a channel, what a placement should cost, or which brands are buying in your niche. Answers come back in seconds, each with a link to the record.
Setup takes one paste. Cancel anytime.