Skip to main content

Teikametrics MCP: Overview

What is MCP?

MCP stands for Model Context Protocol — an open standard that lets AI models like Claude and ChatGPT connect directly to external tools and data sources. If an API connects one piece of software to another, an MCP connects your software to an AI model. Once connected, that tool becomes a data source the model can query in real time, in response to plain-language questions.

The Teikametrics MCP applies this to your marketplace advertising and catalog data, built on our platform's history of processing over $10 billion in marketplace transactions.

What you can use it for

Once connected, you can ask your model plain-language questions instead of building dashboards or exports. A few examples:

  • "Calculate blended contribution margin across my full catalog. Flag any ASIN where ACOS looks healthy but margin is negative."

  • "Compare my top 10 products across Amazon, Walmart, and TikTok Shop. Flag any price-parity risk, stockouts, or aging inventory."

  • "Sponsored Products revenue dropped 18% this week on this ASIN — check catalog changes, inventory, competitor rank, and bid changes, and tell me the most likely cause."

  • "Audit my listings against our brand guidelines and Amazon's latest content policy. Flag anything out of compliance."

Because the model reasons across every connected data source in one conversation, questions that used to require exporting from three systems and doing VLOOKUPs in Excel can now be asked directly, live.

How a query actually works

  1. You ask a question in plain language — for example, "Compare my top 10 products across Amazon, Walmart, and TikTok Shop, and flag any price-parity risk."

  2. The model identifies which of the 12 tools it needs to answer the question.

  3. It queries that data live, scoped to your login's permissions — not from a cached export.

  4. It reasons across the results — merging, comparing, and calculating the way an analyst would.

  5. You get a structured answer, with the option to ask for the underlying formula and inputs so you can verify the work before acting on it.

What tools are available

There are 12 tools, all read-only — this connector is for pulling performance and content data, not managing campaigns.

Advertiser lookup

  • list_advertisers — every advertiser your login can access: name, advertising ID, channel (Amazon/Walmart/TikTok), owning account, country

Campaign-level performance

  • get_campaign_performance — paginated campaign list with impressions, clicks, spend, sales, ACOS, ROAS, CTR, CPC

  • get_campaign_performance_aggregate — one rolled-up summary across all campaigns

Ad group–level performance

  • get_ad_group_performance — paginated ad groups with the same metric set, plus IDs and status

  • get_ad_group_performance_aggregate — rolled-up summary across ad groups

Product ad–level performance

  • get_product_ad_performance — one row per product ad (ASIN/item), with title, ad state, and metrics

  • get_product_ad_performance_aggregate — rolled-up summary across product ads

Keyword / target-level performance (most granular)

  • get_target_performance — targets by text, match type, bid, and metrics

  • get_target_performance_aggregate — rolled-up summary across targets

Catalog / profitability

  • get_product_performance — one row per SKU blending ad + organic data: sales, gross profit/margin, COGS, fees, inventory, TACOS, and more

Listing content & optimization

  • get_listing_optimization_data — Smart Pages status per listing (enrollment, optimization state, whether a content recommendation is ready)

  • get_product_generated_content — current live listing text vs. AI-generated recommendation, plus any user edits, per product

Shared parameters across tools

  • advertising_type: sponsoredProducts, sponsoredBrands, sponsoredDisplay (Amazon only), sponsoredVideo, productSales (TikTok), GMVMaxProduct (TikTok), GMVMaxLive (TikTok)

  • lookback_window: fixed windows only 7d, 14d, 30d, 90d (you cannot query arbitrary custom date ranges)

  • marketplace: amazon, walmart, tiktok (availability varies by tool/advertising_type)

  • Reporting lag by design: windows end 1 day before today for Amazon, 2 days before today for Walmart

Handling your own COGS and financial data

Teikametrics lets you upload your own COGS data, mapped to your own item ID or the marketplace's products in app. By default, the model will use whatever COGS data you've uploaded in app but if you want a specific definition of COGS or margin, you have two options:

  • Give the model instructions in the prompt describing exactly how you want it calculated, or

  • Save those instructions as a reusable skill, so every future conversation uses your definition automatically

Either way, the model reasons against your definition rather than a generic one. You're also not limited to uploaded COGS — the connector exposes raw data points (sales, fees, ad spend, and more), so you can build your own profitability logic from scratch if you'd rather.

Limitations

  • Read-only. There are no write or campaign-management actions in this release. The team is deliberately holding off on write capabilities until audit trails (who made a change, and from where) are built out — that's on the roadmap, not yet available.

  • Fixed lookback windows. You can pull 7, 14, 30, or 90 days of data (depending on the tool), but not an arbitrary custom range.

  • Rate limiting. The only other constraint is rate limiting, in place purely to protect service quality for other customers — you can pull as much data as you need, just not in an unbounded burst.

Best practices

  • Use a paid plan, whichever platform you're on. It keeps your data out of any model training set and gives you more headroom for longer analysis sessions.

  • Opt out of model training in your account settings — a good habit for any proprietary business data you're feeding into an LLM.

  • Use Projects to store your team's common instructions (brand guidelines, your COGS definition, tone, etc.) so every conversation benefits automatically and your whole team gets consistent output.

  • Start with a mid-tier model (we recommend Sonnet for Claude). It's fast, cost-effective, and handles the large majority of MCP queries well. Move to a frontier model only if you're seeing quality issues — frontier models are considerably more expensive and often unnecessary for this kind of analysis.

  • Turn good conversations into skills. Once the model produces an analysis you like, ask it to save the approach as a skill. The next time you ask something similar, it will automatically load those instructions and repeat the process — and you can share the skill with your team so everyone gets the same consistent result. You don't need to know how to write a skill yourself; the model can build it for you from the conversation.

What's on the roadmap

  • Market IQ data — deeper competitive and market signal, to expand beyond your own catalog and ads data

  • Write/action capabilities — currently under review, pending a clear audit-trail design (who made a change, and where)

  • Broader model support — ChatGPT is supported today alongside Claude; Gemini is being evaluated next

Which models does it work with?

At launch, the Teikametrics MCP connects with both Claude and ChatGPT. See the dedicated onboarding article for each platform for setup steps. We're evaluating Gemini support as well and will share timing when we have it.

Questions?

Reach out to contact support if you run into a connection issue.

Did this answer your question?