How to Use the Cruva MCP with Claude Code to Automate TikTok Shop Work

*Guest post by Sohun, founder of Clankers, an operator practice that builds AI-driven TikTok Shop workflows on top of Cruva.*

I run TikTok Shop programs, and for years, the data side meant the same manual grind: exporting spreadsheets, eyeballing creator lists, and copy-pasting numbers into a report every Monday. Connecting the Cruva MCP to Claude Code removed almost all of it. Instead of clicking through dashboards, I describe what I want in plain English, and Claude pulls the data, filters it, and hands back the answer. Below is the exact setup I use.

The short version, for anyone who landed here from a search: the Cruva MCP is a connector that lets an AI model read and act on your live TikTok Shop data. Connect it to Claude Code, and you can find shops, pull creator and GMV data, rank and filter creatives, and even draft outreach DMs, all from a prompt, with Claude writing and running the queries for you. Here is what the MCP is, how to connect it, and the first prompts to run.

What the Cruva MCP Is

MCP stands for Model Context Protocol, an open standard for connecting AI models to live tools and data. The Cruva MCP exposes Cruva's TikTok Shop data and actions as a set of tools an AI can call: searching the marketplace, pulling a shop's products and creators, reading performance metrics, and sending creator messages.

The important word is *live*. Without a data connection, an AI model answers from memory and guesses. With the Cruva MCP connected, every answer traces back to a real number pulled from your shop this minute.

Why Claude Code Is the Right Driver

You can use the Cruva MCP from a chat connector, and for quick one-off questions like "what is this brand's top SKU?", that is the right surface. But most real TikTok Shop work is a small, repeatable pipeline: pull data, apply a couple of filter rules, sort, and render it as something readable.

Claude Code is built for exactly that shape. It writes and runs the queries against the MCP in your terminal, iterates if the data is not quite what you expected, and saves the result to a file you can reuse. When you want the same pull every Monday morning without re-prompting from scratch, Claude Code is the surface you want.

Connecting the MCP

Setup is a one-time step: add the Cruva MCP server to your Claude Code configuration with your Cruva credentials, and confirm the connection is live before you prompt.

Once connected, Claude Code can see the Cruva tools and call them on your behalf.

The Core Cruva Tools

You do not need to memorize these, because Claude picks the right ones from your prompt, but it helps to know the shape of what is available:

• Find and scope a shop: Resolve a shop by name and compute a live date window so runs are reproducible.
• Search the marketplace: Find brands and products by category or keyword.
• Pull creators and performance: Get the creators attached to a brand or product, with GMV, view rates, and ROI.
• Read GMV Max creatives: Pull ad creatives ranked by metrics like the 2-second view rate.
• Score and analyze: Pull performance scores and affiliate data for benchmarking. For sellers who want to understand which of these signals actually predict conversions, TikTok Shop analytics breaks down which metrics matter most before you build automated pulls around them.
• Act: Draft and send creator DMs, and read the inbox.

That last category is why the MCP is more than analytics: it can move from insight to action in the same run.

Want to see the full tool set in action before wiring it into your own workflow? Book a demo to see how Cruva's affiliate discovery, outreach, and performance tracking work together, on the platform or through the MCP.

Your First Five Prompts

The fastest way to understand the MCP is to run it. Try these in order:

1. "Find the [brand] shop and show me its top-selling products over the last 30 days by GMV." This orients you and confirms the connection is live.

2. "Pull the creators driving that top product's GMV, with their view rates and ROI, as a table I can scan." This turns a creator roster into a decision surface.

3. "Find creators in [category] growing more than 50% week over week who are not in the top 100." This is the recruiting pull that surfaces under-the-radar talent.

4. "Rank this brand's GMV Max creatives by 2-second view rate and keep only the ones above 33%." This isolates the creative worth amplifying.

5. "Draft a Spark Ad permission DM for each of these creators, but do not send anything until I approve.' This moves to action, with a human gate—the same outreach engine behind Cruva's TikTok shop affiliate outreach bot, now callable directly from your terminal."

Notice the pattern: you describe the rule, Claude writes and runs it, and you get output you can act on. The skill is defining the rule, not the engineering.

From Prompt to Repeatable Workflow

A single prompt is useful. The real leverage is turning the good ones into workflows you rerun on a cadence. Once you have a pull that works, say the dark-horse creator search, Claude Code can save the script so next week's run is one command instead of a fresh conversation.

That recruiting pull is a good first workflow to make repeatable, and it is the highest-leverage one for most programs, especially once you're ready to move beyond one-off outreach into building a thriving creator network that keeps generating GMV long after the first DM.

Clankers wrote up the full version, covering the growth-rate rule, the top-100 exclusion, the competitor set, and a worked example, in the dark horse creator recruiting workflow. It is the natural second thing to build after your first few prompts.

Frequently Asked Questions

Here are answers to some commonly asked questions:

What Is the Cruva MCP?

It is a connector that exposes Cruva's live TikTok Shop data and actions, including marketplace search, creator and GMV data, performance scores, and creator messaging, as tools an AI model like Claude can call directly, so answers come from real data instead of guesses.

Do I Need To Know How To Code To Use It?

No. You write prompts in plain English describing what you want. Claude Code writes and runs the actual queries against the MCP.

The skill is defining the rule, meaning the growth threshold, the competitor set, and the filter, not writing the script.

Which AI Model Should I Use?

For quick single-metric lookups, a smaller model is fine. For anything involving lots of cross-referencing, like a full competitive benchmark or a multi-step workflow, use the most capable model available. It handles the load noticeably better.

What Should I Automate First?

Start with a repeatable data pull you do every week, like creator recruiting or your weekly report.

Those give the fastest payback and are the easiest to turn into a one-command workflow.