VideoCoreAll assistant connector details

YouTube MCP for ChatGPT

VideoCore's connector puts real YouTube numbers inside ChatGPT. Ask in plain words about your own channel or any public channel and ChatGPT answers from measured data, not from guesswork. It only reads: nothing it does can post to or change your channel.

Prompts to paste

How to connect it

  1. Make a free VideoCore account and connect your YouTube channel. The connector reads your own channel, so it needs one.
  2. In ChatGPT open Settings and then Connectors. Custom connectors are not on every ChatGPT plan, and on some you must first switch on developer mode there; if you do not see the option to add one, your plan does not offer it yet.
  3. Add a custom connector, name it VideoCore and paste this address:https://videocore.uk/api/mcp.php
  4. Choose sign-in (OAuth) if it asks how to authenticate, then press Connect. VideoCore opens: sign in, pick your workspace and press Connect. You are sent back to ChatGPT and it is ready.

There is no key to copy. ChatGPT's connector screen can only use VideoCore's own sign-in, so that is how it connects: you approve it once, and you can disconnect it any time from Settings. A pasted key is for tools that take one, and for code there is the web API.

What you can ask

Diagnose a video diagnose_video

Diagnose why a specific YouTube video is or isn't getting views. Returns YouTube's counts (views, likes, comments) plus figures VideoCore calculates from them (view velocity, engagement, performance vs the channel's OWN recent median, a verdict), and an AI teardown of the thumbnail, title and hook with specific fixes. Use when a creator asks why a video under/over-performed.

Audit a channel channel_audit

Audit any YouTube channel from its public YouTube numbers: subscribers and total views (from YouTube), plus upload cadence, recent median views, best and worst recent videos, and consistency (calculated by VideoCore). Use for competitive intelligence or auditing your own channel.

Find breakout videos on a channel find_outliers

Find a channel's outlier videos — the ones that massively over- or under-performed versus that channel's own recent median. Reveals what topics/formats are actually working right now. Use to find breakout patterns to replicate, and for asks like "the five biggest channels on my topic and which of their videos beat their usual numbers" (call it once per channel).

Rate a niche find_niche

Rate a YouTube niche/topic: demand, competition, a rough RPM estimate and trend, with an overall viability label (all calculated by VideoCore, not YouTube metrics) and the top-performing videos. Use to evaluate whether a niche is worth entering.

What is trending in a niche trending_in_niche

Surface videos gaining serious momentum in a niche. Default: recent uploads (last ~45 days) ranked by view velocity (views per day), to spot formats and topics before they peak. With window "24h": what broke in the niche in the last 24 hours, grouped into stories, each with how many videos and channels are already out on it, when the first went up and the fastest views per hour, plus the creator channel promise when one is saved, so you can say which angle is still open and fits.

Title ideas for a topic keyword_ideas

Generate title/keyword angles for a seed topic, grounded in the title patterns of the real top-performing videos for that topic. Returns the recurring patterns + the real videos they came from.

Compare channels compare_channels

Compare 2-4 YouTube channels side by side: subscribers (from YouTube), recent median views, upload cadence, consistency and recent best video (calculated by VideoCore). Use to benchmark your channel against competitors.

Next video ideas video_ideas

Generate specific next-video ideas for a niche, grounded in what is actually overperforming right now (real top performers + recent momentum). Returns concrete concepts with a title, angle and why it should work.

Check a title check_title

Run VideoCore's measured checks on a YouTube title (the same checks as the public Title Checker): whether it names a person or a situation and not only a topic, whether the words on the thumbnail repeat the title (when the creator gives them), and, for a published video, whether its first thirty seconds deliver the title's key words. Each check is pass, flag or not checked; curiosity, clarity, length and emotional pull are listed as not checked because no code measures them. Never a score. Use before a creator publishes, or to compare title options.

My channel's seasons my_seasons

The creator's own connected channel only. part "months" (default): each calendar month's views a day against the channel's own months around it, by year, and which months repeat as above or below usual in every year measured (needs a year of history; says so when there is not enough). part "titles": the channel's own uploads with a year or a seasonal phrase in the title (Christmas, summer, a month name) against its untagged uploads, by publish month; give "title" to see the channel's own videos with the same markers. Its own history, not a general rule, and not a forecast. Use when a creator asks when to publish, about seasonal topics, or whether to put a year in a title.

Where viewers left my video my_retention

The creator's own connected channel only, long videos (three minutes or more). For one video (newest by default): the share of people who started it still watching at 0:30, a quarter, halfway, three quarters and near the end, beside the channel's own usual at the same points; the steepest drop in the opening; the three steepest drops after it; and up to three places where more people were watching than a moment before (they rewound to it or skipped ahead to it; the curve cannot say which). When VideoCore holds the captions, each drop and rise carries what was being said there. From YouTube Analytics curves VideoCore already stores; no lookup is used. Use when a creator asks why people stop watching, or for edit notes for the next cut. Write notes from these times and words only; the reason people left is the creator's judgement.

My VideoCore projects my_projects

List the videos (projects) in the key holder's own VideoCore workspace, newest first, with whether each has a script.

What to know first

Also

Using Claude instead? VideoCore for Claude. Using code or a spreadsheet? The VideoCore API has the same lookups. Something not working? Use Report a problem inside VideoCore.