What is the Teikametrics MCP?
A connector that lets you point Claude or ChatGPT directly at your Teikametrics data — ARI Catalog, Ads, and profitability data across Amazon, Walmart, and TikTok Shop — so you can ask questions in plain language instead of building dashboards or exports.
Which platforms does it work with?
Claude and ChatGPT, both as of this release. Gemini support is being evaluated, with no timing to share yet.
Do I need Claude or ChatGPT specifically, or can I use whichever I already have? Either works — both connect to the same underlying data and tools. Use whichever your team already works in.
Is my data used to train Claude or ChatGPT's models?
Not if you're on a paid plan with training turned off in your account settings, which we recommend regardless of which platform you use. This is a setting in your own Claude or ChatGPT account, not something Teikametrics controls — worth confirming on your end before connecting proprietary data.
Can the MCP make changes to my campaigns, listings, or bids?
No — everything in this release is read-only. The MCP can pull and analyze data, but it can't create, edit, or delete anything in your account. We're intentionally holding off on write/action capabilities until we've built out a clear audit trail (who made a change, and from where), so this is a deliberate design choice, not a current-version gap.
How much data can I pull?
As much as you need. The lookback windows are fixed per tool (7, 14, 30, or 90 days — you can't query an arbitrary custom range), but there's no cap on how many queries you run. The only constraint is rate limiting, which exists purely to protect service quality for other customers, not to limit your usage.
Why can't I query a custom date range?
Lookback windows are fixed by design in this release (7d/14d/30d/90d, depending on the tool). If you need a specific window not currently offered, let your account team know — it's useful input for the roadmap.
Why does my data look like it's missing the most recent day or two?
That's expected reporting lag, not a bug: Amazon data reflects through 1 day before today, and Walmart through 2 days before today.
What's the source of truth for COGS?
Whatever you've told it to be. Teikametrics lets you upload your own COGS data mapped to your item ID or ASIN, and the model will use that by default. If you want a specific calculation or definition of COGS or margin, just describe it in a prompt or save it as a skill — the model will reason against your definition instead of a generic one. You can also skip uploaded COGS entirely and build your own profitability logic from the raw sales, fee, and ad-spend data points the connector exposes.
Which model should I use — Sonnet, Opus, GPT-4, etc.?
Start with a mid-tier model (we recommend Sonnet on Claude). It's fast, cost-effective, and handles the large majority of MCP queries well. Only move to a more expensive frontier model if you're seeing quality issues — for most of these use cases, frontier models are overkill and considerably more expensive.
What's a "skill," and do I need to know how to write one?
A skill is a saved set of instructions that a model automatically loads when it recognizes a similar task, so you don't have to re-explain your preferences every time. No, you don't need to know how to write one — once you get a good result from a conversation, just ask the model to save that approach as a skill, and it will build it for you. Skills can also be shared with your team so everyone gets consistent results.
Can I share a skill or Project with my team?
Yes — both Claude and ChatGPT support Projects for storing shared instructions (brand guidelines, your COGS definition, etc.), and skills can be shared the same way, so your whole team benefits from work one person has already done.
Why is ChatGPT setup more involved than Claude's?
ChatGPT requires a User-Defined OAuth Client, which means generating a callback URL that our team has to manually register before the connection activates. Claude's custom connector doesn't require that back-and-forth. We're looking at ways to streamline this over time.
I set up the ChatGPT connection and got an error when I tried to sign in. Is something broken?
No — this is expected the first time, before we've registered your callback URL. Once we confirm registration, go back to your plugin list and hit Connect again.
Is there a cost to use the MCP?
The MCP connector itself is part of your Teikametrics access. You're still responsible for whatever Claude or ChatGPT plan and usage costs you incur on the model side.
When does early access open, and how do I get in?
Sign up at mcp.teikametrics.com. Access has been rolling out to the waitlist on a roughly 1–2 week timeline after signup.
What's coming next?
Market IQ data (deeper competitive signal), write/action capabilities (pending audit-trail design), and possibly Gemini support. If there's a specific capability or data source you'd like to see, tell your account team — that feedback goes directly into the roadmap.