Use Case
TikTok Data for Claude: Read Tracked Accounts, Posts and Comments Without Leaving the Chat
Give Claude the TikTok accounts, saved posts, comments and metrics your Canvas tracks, and run new collection only at an approved price.
By Bell Chen, founder of Superdirector — how the analysis works.
Reference visuals
Frames and thumbnails behind this guide
The best TikTok research I have seen rarely starts with views. It starts with the comments: the question a viewer asked under a competitor video, the third time someone requested a tutorial, the objection nobody answered. That is evidence Claude reads well, as long as it can actually get to it in a form that says where each comment came from and when it was collected.
Claude can call tools on remote servers through the Model Context Protocol (modelcontextprotocol.io). Connected to a Canvas that tracks your TikTok accounts and the creators you study, it can read the saved posts and comments directly. This page walks through that arrangement, including what it will not do.
What Claude can reach on TikTok through a Canvas
The Canvas holds TikTok sources by role: your own account, competitor accounts, inspiration creators whose formats you study, and hashtag and keyword sources for your niche. Every completed read saves the posts it returned with captions and metrics, the comments when comment collection was part of the read, and account-level metrics. Claude reads those records through native MCP tools and reports the coverage it had.
Comments are the part most worth connecting. Claude can page through the saved comments for one read, group them by what viewers are asking, and quote the exact comment beside each theme. Because each read is dated, Claude can also tell you whether a theme is new or has been there for months.
Covers matter on TikTok because the first frame decides whether a grid visitor taps. Open the saved posts on the Canvas to compare framing and on-screen text across creators; the connection returns no images, so Claude works from captions, hooks and comments. Watching the video, reading its frames or transcribing it takes a video analysis report.
A comment-led research session in Claude
| Stage | When / duration | Tools | Deliverable |
|---|---|---|---|
| 1. Connect Claude and confirm the tools are live | about 5 minutes | Claude connector settings or Claude Code MCP config | account status and a list of your Canvases |
| 2. Pick one source and read its comments | 10 to 15 minutes | knowledge_inventory, read_source_evidence | comment themes with quoted examples and the read date |
| 3. Compare openings on the Canvas | 5 minutes | open_canvas | a short note on first-frame patterns |
| 4. Widen with a quoted run only if needed | minutes to approve, longer to complete | quote_module_run, run_module | a new committed read |
| 5. Turn themes into a plan and save it | 15 minutes | save_artifact | a script or content plan in Canvas Files |
Connect Claude and confirm the tools are live
Add the remote server, sign in when the browser opens, and approve the OAuth grant (modelcontextprotocol.io). Then ask Claude to check your account status and list your Canvases. Both are free and confirm the connection is real before you ask anything substantive.
Pick one source and read its comments
Start narrow: one competitor, its most recent completed read. Ask Claude to list the saved posts, then read the collected comments under the five posts that drew the most replies, and group them by what the viewer wanted. Ask it to quote each example and name the post it came from.
If some posts have no collected comments, Claude should say so rather than infer them. That gap is information too: it tells you which read to widen next.
Compare openings on the Canvas
For the posts behind the strongest comment themes, ask Claude for the Canvas link and look at the covers there: the framing, the on-screen text and the subject. The connection returns no images, so the visual read is yours; Claude can set it beside the captions and comments it read.
Widen with a quoted run only if needed
If the theme deserves more evidence, ask Claude to quote a read of another creator or a hashtag source. Approve the price in the chat; Claude starts the run, keeps the run ID, and reads the result when it commits.
Turn themes into a plan and save it
Ask Claude to write a script that answers the most repeated unanswered question, citing the comments that prompted it, and save it to the Canvas. You film and post it yourself; the saved plan keeps the evidence attached for whoever edits it.
Common mistakes
Asking for trends with no date. "What is trending in my niche" has no answer without a time window. Ask what the saved reads from the last two weeks show, and let Claude tell you which sources are older than that.
Counting comments instead of reading them. Volume tells you a post provoked a reaction. The words tell you what to make next. Ask for quoted examples beside every theme.
Assuming Claude watched the video. It read the caption, the metrics and the comments. If the edit is the point, run a video analysis report.
Where Superdirector fits
Superdirector is the Canvas Claude connects to. Connecting and inventory checks are free; paid plans start at $29 per month, and new collection is metered in credits with a quote and your approval before every run. A verified account gets a one-time 20-credit MCP starter inside the AI app, at most one account check and one X text draft. The MCP starter is not a trial of the plan; Starter has its own 7-day free trial. Your AI cannot post.
The same connection carries Instagram (Instagram MCP Server: Your Tracked Instagram Data Inside Claude, ChatGPT and Codex) and works the same way from ChatGPT or Codex (Social Media MCP for ChatGPT and Codex: One Canvas, Four Platforms, Any MCP Client).
Disclosure: I build Superdirector. Tool names and limits reflect the MCP server as of the date above and can change; Claude reads the current capabilities before promising an action. TikTok is a trademark of ByteDance, and Claude is a product of Anthropic; neither is affiliated with this page.