How to Connect TikTok Data to Claude (MCP Setup Guide, 2026)

The problem with researching TikTok in a chat window

You have an AI assistant that is good at spotting patterns, and a pile of TikTok data it cannot see. So you do what everyone does: open the analytics tool, copy some numbers, paste them into the chat, describe what you are looking at, and hope the context survives.

It mostly does not. You paste view counts without the creator's baseline, so the assistant cannot tell a breakout from a normal post. You paste a transcript without the scene timings, so it cannot tell you why the hook worked. You end up doing the analysis yourself and using the assistant as a thesaurus.

MCP fixes this by letting the assistant fetch the data itself. You ask "what's breaking out in crypto this week and what hooks are they using", and it pulls the videos, reads the teardowns, and answers — with the baselines and timings intact, because it fetched them rather than being told about them.

This guide shows the setup, then the prompts that are actually worth using.

What you need first

• Claude, on a plan that supports custom connectors.
• A FYPNow workspace on a paid plan. API and MCP access are included on every paid tier at no extra cost — there is no separate add-on — but they do require an active subscription. A free trial counts.

That is it. No API key, no config file, no terminal.

Step 1: Add a custom connector in Claude

In Claude, open Settings → Connectors and choose Add custom connector.

Step 2: Paste the server URL

https://api.fypnow.com/mcp

Leave the authentication fields blank. This is the part people get wrong — they assume an MCP server needs an API key pasted into a header. Claude's custom connectors use OAuth, and the server advertises that: Claude fetches a discovery document from the server, finds the authorization server, registers itself, and starts the flow. Nothing for you to fill in.

Step 3: Approve it, and pick your workspace

A FYPNow window opens. Two choices on it, and both matter:

Which workspace the connection acts on. Every connection is bound to exactly one. If you belong to more than one, pick deliberately — a connection bound to the wrong one will return a permission error on every single tool call, with nothing in Claude explaining why. Read-only, or read and write. Read-only can browse niches, look up existing analyses, list breakouts and check usage. It cannot analyze a new video, create a niche, search trends or generate a script — those tools are not even shown to it. Choose read-only if you are just exploring; you can reconnect with write access later.

Approve, and the tools appear in Claude immediately.

Step 4: Ask it something real

The assistant picks the tools itself. Prompts worth starting with:

Find what is working in a niche

What are the top videos in the Crypto niche right now, and what hooks do they use?

This browses the niche catalogue, ranks by how far each video outran its creator's own median, and pulls the hook analysis for the top few.

Tear down one specific video

Analyze https://www.tiktok.com/@someone/video/123 — how long is the hook, and where are the scene cuts?

Returns hook type and text, scene count and timings, the transcript with timestamps, retention signals and a read on why it travelled.

Check your own tracked creators

Which of my tracked creators had a video break out this week?

Breakout detection compares each video against a 14–60 day peer baseline, so this is outperformance rather than raw views.

Find more like one that worked

Find videos similar to this one in my library.

Vector search over the analyses, so "similar" means similar in structure and content, not matching hashtags.

Write from your own winners

Write me a script in the format that performed best on my account last month.

A realistic note on speed

Two things are fast and one is not.

Reads — browsing niches, looking up an existing analysis, listing breakouts — come back in well under a second.

A fresh AI teardown takes roughly a minute per video, because it runs a video model over the actual footage. If you ask for hooks across five videos that have never been analysed, expect a few minutes and an assistant that sits there polling. That is the honest number. Videos already analysed return instantly.

Worth knowing so you frame the request well: ask about a handful of videos, not thirty.

What it does not do

• It is not a scraper you can point anywhere. A connection acts on one workspace, against that workspace's plan limits.
• It cannot post. FYPNow reads and analyses; publishing stays with you.
• It does not see your private TikTok account data. It analyses public videos, plus whatever is in your own workspace.

Revoking it

Open Settings → API in FYPNow. Connected apps lists every connection you have authorised, which workspace each acts on, the scope, and when it was last used. Revoke cuts it on both sides — the token stops working and the API refuses it on the very next request.

Does this work with ChatGPT?

The OAuth flow ChatGPT's app directory requires is live, but FYPNow is not yet listed in that directory, so there is no one-click install there. A ChatGPT custom action can reach the same data over the REST API today — see how to connect TikTok data to ChatGPT.

Where to go next

• The MCP server page — all 18 tools, what each one costs
• Setup guide in the docs — the same steps, plus bearer-key clients like Cursor and VS Code
• Find viral TikTok videos in 4 steps — the whole research flow, agent-first