TIKTOK SHOP · JUN 10, 2026 · 7 MIN
How to Automate a TikTok Shop Competitive Analysis with Claude and the Cruva MCP
Connect a TikTok Shop data tool like Cruva to Claude over MCP, pull four months of GMV, creator, and conversion data for a set of competitors, and Claude builds a benchmark scorecard with recommendations you can send straight to a client. Here is the exact workflow, step by step.
A TikTok Shop competitive analysis used to mean a day of exporting spreadsheets from an analytics tool, stitching them together, and writing up the comparison by hand. With an MCP connection between Claude and a TikTok Shop data source, the whole thing runs from a single prompt: Claude pulls the marketplace data directly, benchmarks the brand against a competitor set, and returns a readable scorecard with recommendations.
The short answer, for anyone who landed here from a search: connect a TikTok Shop outreach or analytics tool (I use Cruva) to Claude via MCP, prompt Claude to find the brand through the tool's marketplace search, pull units sold, GMV, price, and creator data over the last four months, benchmark against relevant competitors on creator recruitment, creator retention, and video conversion, and output a scorecard with high-level summaries and recommendations. Then export the result as HTML so it is presentable to a client.
The rest of this piece walks through each step, the reasoning behind the metric choices, and a worked example from a real run.
Step 1: Connect Your Data Source
The analysis is only as good as the data behind it, so the first move is connecting a tool that has direct TikTok Shop marketplace data into Claude. I use Cruva, but other TikTok Shop tools with MCP support, Euka for example, work the same way. What matters is that the connection is live inside Claude before you prompt, so Claude is querying real marketplace data rather than reasoning from memory.
This is the part people skip and then wonder why the output reads like a generic strategy essay. Without a connected data source, Claude has no units, no GMV, no creator counts. With one, every claim in the scorecard traces back to a real number.
Step 2: Make Claude Find the Direct Data Source
The first instruction in the prompt is about data discipline: tell Claude to use the tool's marketplace search for brands, so it locates the brand's actual listing data rather than approximating from videos or general search.
From there, the prompt should make Claude:
- Pull the core commercial data: units sold, GMV, price points, and the creator roster attached to the brand
- Build a set of relevant competitors to benchmark against, brands in the same category at a comparable price point
The competitor set is what turns raw numbers into analysis. A 6 percent figure means nothing on its own. Against four competitors in the same category, it becomes a diagnosis.
Step 3: Benchmark the Affiliate Metrics That Actually Matter
GMV tells you the score, but it does not tell you why a brand is winning or losing. On TikTok Shop, the why almost always lives on the affiliate side. These are the metrics I have Claude benchmark, and what each one signals:
| Metric | What it signals |
|---|---|
| Total creators recruited | Top-of-funnel strength: is the product attractive enough that creators enroll? |
| Creator retention over time | Whether the brand keeps creators posting after the first video: follow-up, community, ongoing incentives |
| Conversion on creator videos | Content quality: creative briefing, selling points, coaching, community support |
| Overall GMV | The validation layer: confirms whether the affiliate machine is actually producing sales |
The interplay between these is where the insight comes from. A brand can look healthy on recruitment and still be leaking everywhere downstream, which is exactly what the worked example below shows.
Step 4: Why a Four-Month Window
Run the analysis over the last four months. A 30-day snapshot is noisy: one viral video, one promotion, one seasonal spike can dominate the picture. Four months gets you over a full quarter, which is long enough to separate trajectory from noise and to stop variables like seasonality from confounding the comparison.
If a brand's retention looks strong in the last 30 days but weak over four months, that distinction matters, and a shorter window would have hidden it.
Step 5: Define the Scorecard Before You Run
Tell Claude exactly what the output should look like before it runs. I ask for a scorecard format: each metric scored against the competitor set, a high-level summary per section, and recommendations at the end. The reason is practical. A scorecard with summaries can be sent to a client or a teammate as-is. A wall of analysis cannot, and you end up rewriting it by hand, which defeats the point of automating.
Which Model Should You Use?
I run this on Opus 4.8. The analysis involves a lot of assumptions, computations, and cross-referencing across many different data points pulled from the tool, and the larger model handles that load noticeably better. For a quick single-metric lookup a smaller model is fine, but for a full competitive benchmark, use the most capable model available.
A Worked Example: Diagnosing a Chocolate Protein Brand
I ran this workflow on a chocolate protein brand I thought was doing something interesting. Here is what came back, and how the metrics combined into a diagnosis:
- Recruitment was strong. The brand scored among the highest in the competitor set for creator recruitment in the last 30 days. Lots of creators enrolling at the top of funnel.
- Retention was weak. Creator retention came in at 6 percent, low against the benchmark set. Creators post once and disappear.
- Video conversion was the lowest in the set. The creator videos that did go up converted worse than every competitor.
Read those three together and the diagnosis writes itself. If recruitment is high but retention and conversion are both low, the problem is not recruitment. The product is clearly attractive to creators, that is why so many enroll. What is missing is the layer after enrollment: follow-up, coaching, selling points, support. Creators are not being nurtured into posting again, and the content they do post is not briefed well enough to convert. Sales were lower than the leaders in the set, but with that much top-of-funnel volume, this is a nurturing and communication problem, not a demand problem.
The recommendations that fell out of the scorecard:
- Recruit for post rate, not just enrollment. Screen creators in advance and prioritize the ones with a history of actually posting.
- Build a follow-up and coaching motion after enrollment, so creators get the selling points and support that competitors are clearly providing.
- Fix the briefing layer to lift video conversion. We cover how to scale that without losing brand control in the piece on AI-generated affiliate briefs.
That is a real strategic read, produced from one prompt against live marketplace data. I still do my own pass over the numbers before anything goes to a client, the data is right there to check, but the heavy lifting of pulling, benchmarking, and structuring is done.
Step 6: Export It So a Client Can Read It
The last instruction in the prompt: make the whole analysis exportable as HTML. An HTML artifact plugs straight into a design tool, Canva or similar, so the scorecard becomes a presentable client deliverable instead of a chat transcript. This is the difference between an analysis you did and an analysis you can send.
If you want to take presentation further, the same MCP pattern extends to design tooling. We covered that workflow in producing TikTok Shop creative at scale with Claude and the Figma MCP.
Frequently Asked Questions
Do I need to write code to do this?
No. The setup is connecting an MCP-enabled tool to Claude and writing a clear prompt. The skill is knowing which metrics to ask for and how to structure the scorecard, which is operator knowledge, not engineering.
What does the prompt actually need to include?
Five things: use the marketplace brand search to find the direct data source, pull units, GMV, price, and creators, build a relevant competitor set, benchmark recruitment, retention, video conversion, and GMV over the last four months, and output a scorecard with summaries and recommendations.
Why focus on affiliate metrics instead of just sales?
Because on TikTok Shop, sales are the output of the affiliate machine. Recruitment, retention, and video conversion tell you which part of the machine is broken. GMV alone tells you that something is broken without telling you what.
Can I use a tool other than Cruva?
Yes. Any TikTok Shop analytics or outreach tool with an MCP connection and marketplace-level brand data works. The workflow is tool-agnostic; the metric framework is what carries.
How often should I rerun the analysis?
The four-month rolling window makes this natural to rerun monthly. The trajectory between runs, retention improving, conversion closing the gap to competitors, is often more useful than any single snapshot.
If you run a TikTok Shop program and want this kind of analysis wired into your own operation, get in touch.