OutlierKit MCP Server: YouTube Outlier Research Inside Claude
The OutlierKit MCP server is live. It is a remote connector, so there is nothing to install. You paste one URL into Claude, sign in with Google, and Claude gets ten read-only YouTube research tools. It can search our outlier index, find channels that look like a seed channel, pull a transcript or a comment thread, and check what people actually search for. All of it happens in the chat, with no exports and no copy and paste.
The Short Version
Connector URL: https://outlierkit.com/mcp Add it once in Claude and the tools work in every chat.
Sign in: OAuth 2.1 through WorkOS AuthKit. Use the Google account tied to your OutlierKit login. There is no API key to copy.
What you need: a paid OutlierKit plan, Pro or higher. MCP access is included at no extra cost.
What it costs to run: 1 credit per tool call, from the same balance the app uses. The one exception is outliers_refresh, which costs 5 because it queries YouTube live.
Who this is for: creators who research inside Claude, agency teams doing client work, and developers wiring YouTube data into their own agents.
Want a walkthrough before you wire it into client work?
Book a 30-minute demo with the founder. We'll connect it live against your niche and cover multi-seat and platform pricing.
One remote connector, ten read-only tools, and the same credit balance your OutlierKit app already uses
Connect It to Claude in Two Minutes
OutlierKit runs a remote MCP server, not a package you install. You add the URL once, sign in with Google, and the tools are available in every Claude chat from then on.
Connector URL
https://outlierkit.com/mcp
Claude (web and desktop)
- 1Open claude.ai, then Settings, then Connectors.
- 2Click Add custom connector.
- 3Paste https://outlierkit.com/mcp and click Add. Leave any auth fields blank.
- 4Click Connect. A WorkOS window opens. Sign in with Google and approve.
- 5The connector shows as Connected with its tools listed. Enable it in a chat from the tools menu and ask away.
Use the Google account tied to your OutlierKit login. That is how the connector matches your plan and credits.
Claude Code
One command, then approve the OAuth prompt:
claude mcp add outlierkit --transport http https://outlierkit.com/mcp
Other MCP Clients
Any client that supports remote MCP servers over HTTP with OAuth can point at the same URL. Clients that only run local stdio servers cannot connect, because there is no local package to run. If your tooling needs a key-based integration instead, the OutlierKit REST API reaches the same data with a bearer token.
Requirements
- ✓A paid OutlierKit plan, Pro or higher. MCP access is included.
- ✓A Google account tied to your OutlierKit login, for the sign-in step.
- ✓Credits on your balance. Every tool call spends from the same pool the app uses.
Full connector reference: outlierkit.com/mcp/docs
The Ten Tools
Every tool is read-only and returns structured JSON. Most cost 1 credit per call. Claude picks the right one and chains calls on its own, so you can ask a research question in plain language and let it work out the sequence.
Outliers
outliers_search1 creditSemantic search across our indexed outlier videos. Returns the most relevant outlier videos plus their parent channel context. Fast, because it does not call YouTube.
outliers_refresh5 creditsFetches fresh videos live from YouTube, ranks them by outlier score, and returns them in the same shape as outliers_search. Use it when you need more results or newer ones. Semantic similarity is null on this path. The fetched outliers are saved to the index in the background, so later searches can find them. Pro plan or higher.
Channels
channels_search1 creditSemantic search across our indexed outlier channels. Returns matching channels plus their top outlier videos.
channels_similar1 creditGive it a seed channelId and it returns the most semantically similar channels in our index. Set sizeSimilarity to true to also sort by subscribers and average views per video, which finds channels in the same weight class as the seed.
channels_get1 creditFetch one channel by internal id or YouTube channelId. Serves cached data when it is fresh, otherwise fetches live and refreshes the cache. A source field tells you which path it took.
channels_videos1 creditRecent uploads from a channel. Always fetched live from YouTube, so the list is current.
Videos
videos_get1 creditFetch one video by internal id or YouTube videoId. Cache first, live fallback. For videos we have not seen before, the outlier score is computed in the background and shows up on later requests.
videos_transcript1 creditFetch a video transcript. Cached on the first fetch, since transcripts do not change, so repeat requests are fast.
videos_comments1 creditComments for a video. Always fetched live from YouTube and never cached.
Keywords
keywords_research1 creditExpand seed keywords into a ranked list of related YouTube search keywords with monthly search volumes. Results are deduplicated and sorted by volume. Use minVolume and limit to narrow them.
Cached versus live. outliers_search and videos_transcript read from our index, so they are fast. channels_videos and videos_comments always hit YouTube, so they are current. channels_get and videos_get try the cache first and fall back to live. Knowing which is which tells you when a result might lag reality.
How OutlierKit MCP Differs from YouTube Transcript MCP Servers
There are over 40 YouTube MCP servers in community directories, and nearly all of them do the same thing. They wrap yt-dlp or the YouTube Data API v3 and hand back transcripts, comments, and basic metadata. That is genuinely useful, and it is free. What it cannot give you is an opinion about which videos matter. OutlierKit serves the same primitives plus its own analytics layer on top: the outlier scores our engine computes, a semantic index you can search in plain language, and keyword volume data.
Community servers return what YouTube publishes. OutlierKit returns that plus the scoring and search layer we built on it.
| Capability | Transcript MCP Servers | OutlierKit MCP | Note |
|---|---|---|---|
| Video transcript extraction | ✓ | ✓ | videos_transcript, cached on first fetch |
| Video metadata (views, likes, duration) | ✓ | ✓ | videos_get, cache first with a live fallback |
| Video comments | ✓ | ✓ | videos_comments, always live |
| Outlier scores computed per video | ✗ | ✓ | Our engine scores against channel baseline |
| Semantic search over an outlier index | ✗ | ✓ | Ask in plain language, not by video ID |
| Similar-channel discovery by embedding | ✗ | ✓ | channels_similar, optional size matching |
| Live outlier refresh ranked by score | ✗ | ✓ | outliers_refresh, 5 credits |
| Keyword volumes for YouTube search | ✗ | ✓ | keywords_research, sorted by volume |
| Works without a Google Cloud project | ✗ | ✓ | No YouTube Data API quota to manage |
| Free to run | ✓ | ✗ | OutlierKit needs a paid plan and spends credits |
Honest take on when you need both. Because OutlierKit serves transcripts and comments itself, most YouTube research runs fine on this connector alone. Keep a free transcript server around if you work with videos outside our index a lot, or if you want an unmetered path for bulk caption pulls. Credits are worth spending on the scoring and search layer, not on captions you could fetch for nothing.
Evaluating OutlierKit MCP for production use?
Book a 30-minute demo with the founder. We'll run the tools live against your use case and cover multi-seat and platform pricing.
What Stays in the OutlierKit App
The connector covers the research primitives. The deep analysis modules still run in the app, where they can take their time over a whole niche. That split is worth knowing before you plan a workflow around Claude.
| Job | Where it runs |
|---|---|
| Search scored outliers by topic | MCP connector |
| Find channels similar to a seed | MCP connector |
| Read a transcript or comment thread | MCP connector |
| Keyword volumes for titles | MCP connector |
| Niche-wide competitor mapping | Competitor Studio, in the app |
| Audience psychographic segments | Competitor Studio, in the app |
| Sponsor landscape and monetization | Competitor Studio, in the app |
The workflow that gets the most out of both: run a Competitor Studio scan in the app for the strategic picture, then keep Claude open with the connector on for everything you want to chase afterwards. Here is what the app-side scan adds.
Niche mapping: thousands of competitors identified from one seed channel
Audience psychographics: segment-level drivers and pain points
Sponsor intelligence: landscape mapping across the entire niche
Monetization mapping: revenue funnels and strategy breakdowns
Example Prompts Once the Connector Is On
You do not name the tools. Claude reads what is available and picks. These are the shapes of question that work well, with the tools each one tends to reach for.
outliers_search"Search OutlierKit for outlier videos about personal finance for people in their twenties. Look at the top 10, then tell me what the titles and framings have in common."
outliers_refresh"The indexed results for AI coding tools feel thin. Run a live refresh on that topic, then show me the highest outlier scores from the last month."
channels_similar"Take this channel as a seed and find the most similar channels in OutlierKit, matched on size as well. For each one, show me its top outlier video."
videos_get, videos_transcript, videos_comments"Pull this video, its transcript, and its comments. Tell me how the first 30 seconds is structured. Then tell me what the comments say people wanted more of."
keywords_research"Expand "faceless youtube channel" into related keywords with at least 1,000 monthly searches. Then write 10 title options for the three highest-volume ones."
channels_videos, outliers_search"Pull the last two weeks of uploads from these five channels. Flag anything that reads like an outlier, then check OutlierKit for similar breakouts in the same niche."
A tip that saves credits: ask for the search first, look at what comes back, then ask for transcripts or comments on the two or three videos worth reading. Asking for everything up front spends credits on videos you will skip.
Use Cases by Audience
🏢YouTube Agencies
- •Client intake: run the Competitor Studio scan in the app, then keep Claude open to chase every thread the scan surfaces without leaving the chat
- •Weekly check-ins: pull recent uploads from each client's competitor set and ask Claude what broke out and why
- •Report drafting: the data arrives as JSON in the same chat where you write the report, so the numbers in the deck are the numbers from the tool
- •Pitch prep: research a prospect's niche in the ten minutes before the call, not the day before
📈Businesses Using YouTube for Lead Generation
- •Niche validation: before committing budget to video, check whether anything in the space is actually breaking out
- •Content planning: pair outlier search with keyword volumes, then have Claude turn the overlap into a calendar
- •Competitor tracking: watch what your competitors ship each week and how their audience reacts in the comments
- •Comment mining: pull comments on the breakout videos in your space to find the questions your buyers keep asking
💻Developers Building YouTube AI Tools
- •No YouTube quota to manage: you skip the Google Cloud project, the API key rotation, and the daily unit budget
- •Custom agents: point your own MCP-capable harness at the same URL and give it YouTube research as a capability
- •Prototype in chat, ship on the API: work out the tool sequence in Claude, then reimplement it against the REST API once it is stable
- •Model-agnostic data layer: the connector outlives whichever model you are using this quarter
Credits, Plans, and Access
MCP access is included on every paid plan, Pro and above. There is no separate MCP fee and no add-on to buy. What you are really budgeting is credits, because each tool call spends one from the same balance the OutlierKit app uses.
| Call | Cost | Why |
|---|---|---|
| Nine of the ten tools | 1 credit | Served from our index or a cheap live lookup |
| outliers_refresh | 5 credits | Queries YouTube live, scores the results, and writes them back to the index |
Pro Plan · MCP included
$49/month
or $24.9/mo billed annually
- ✓500 monthly research credits
- ✓All ten MCP tools, including outliers_refresh
- ✓REST API access as well
- ✓Roughly 500 tool calls a month
Max Plan · MCP included
$199/month
or $83/mo billed annually
- ✓2,000 monthly research credits
- ✓All ten MCP tools, including outliers_refresh
- ✓REST API access as well
- ✓Headroom for agency and multi-client work
Both plans also include the REST API, which reaches the same data with a bearer token. Current prices and your credit balance live in the OutlierKit app.
Using OutlierKit MCP with n8n and Automation Tools
The connector is built for a person talking to an agent. For anything on a schedule, use the OutlierKit REST API with n8n or similar. HTTP Request nodes give you retries, error branches, and control over where the data lands, which is what a weekly competitor report or a daily outlier alert actually needs.
Frequently Asked Questions
Is the OutlierKit MCP server available now?+
Yes. It is live. It is a remote MCP connector at https://outlierkit.com/mcp. You add that one URL in Claude, sign in with Google, and the ten tools are available in any chat. There is no download, no npm package, and no API key to paste.
What is an MCP server?+
MCP (Model Context Protocol) is an open standard created by Anthropic. It lets AI assistants call external tools through one shared interface. An MCP server publishes a set of tools the assistant can use. OutlierKit's server publishes ten YouTube research tools. Once it is connected, Claude can search our outlier index, pull a transcript, or check keyword volumes on its own. You never copy or paste the data by hand.
What plan do I need for MCP access?+
Any paid OutlierKit plan, Pro or higher. Pro is $49/month or $24.9/month billed annually. Max is $199/month or $83/month billed annually. MCP access is included at no extra cost. One tool, outliers_refresh, is Pro plan or higher because it queries YouTube live.
Do I need an API key to use the MCP server?+
No. The connector uses OAuth 2.1 through WorkOS AuthKit. You sign in with Google using the account tied to your OutlierKit login, and consent once. OutlierKit reads only your email address, so it can match your account and apply your plan and credits. The REST API, which does use an API key, is a separate way to reach the same data.
How many credits does each MCP tool call cost?+
Most tools cost 1 credit per call. The exception is outliers_refresh, which costs 5 credits because it fetches fresh videos live from YouTube. Credits come from the same balance the OutlierKit app uses, so a research session in Claude draws down the same pool as a session in the web app.
Does the connector get access to my YouTube channel?+
No. It reads your email address to match your OutlierKit account. It does not touch your YouTube channel, your Google account, or any private Google data. Every tool is read-only, and all of them call the OutlierKit API on your behalf.
Can I use it with Claude Code, Cursor, or another MCP client?+
Claude Code is supported directly: run claude mcp add outlierkit --transport http https://outlierkit.com/mcp. Any other client that supports remote MCP servers over HTTP with OAuth can connect to the same URL. Clients that only support local stdio servers cannot, because there is no local package to run.
Can I use OutlierKit MCP with n8n?+
The connector is built for interactive AI agents. For scheduled workflows in n8n, use the OutlierKit REST API with HTTP Request nodes. You get more control over scheduling, retries, error handling, and where the data lands.
How does OutlierKit MCP differ from youtube-mcp or mcp-youtube-analytics?+
Community YouTube MCP servers wrap the public YouTube Data API and yt-dlp, so they return captions and basic metadata. OutlierKit returns its own analytics layer on top of that: a proprietary index of outlier videos and channels, the outlier scores our engine computes, embeddings-based similarity search, and keyword volume datasets. OutlierKit also serves transcripts and comments, so for most YouTube research you do not need both.
Does the MCP server run a full Competitor Studio scan?+
No. Competitor Studio, with its audience psychographics, sponsor mapping, and monetization breakdowns, runs in the OutlierKit app. The MCP tools cover the research primitives: outlier search, live outlier refresh, channel and video lookups, similar-channel discovery, transcripts, comments, and keyword research. A practical split is to run the deep scan in the app, then use Claude to interrogate the niche around it.
The connector is live. Add it in two minutes.
Paste https://outlierkit.com/mcp into Claude's custom connector setting, sign in with Google, and start researching. Included on every paid plan.
