TikTok Ads + ChatGPT: 13 analyses you run in natural language using Kondado's MCP

Asking "which are my 10 TikTok Ads campaigns with the best ROAS in the last 30 days?" straight to ChatGPT or Claude works when your TikTok Ads account is connected to Kondado. Kondado replicates your TikTok Ads tables — 17 tables with 80+ metrics, from spend and impressions to video_views_p100, complete_payment and landing_page_view_rate — into Via Kondado, Kondado's managed destination, and the MCP server at https://mcp.kondado.io/mcp is the read-only channel through which ChatGPT and Claude read that base. The assistant never talks to the TikTok API: it reads the tables already materialized in Via Kondado, writes the query in KSQL and returns the analysis in seconds, straight from the chat — no SQL, no BigQuery.

Most people running TikTok Ads never query these tables, because that used to require SQL, knowledge of the hierarchy (advertiser → campaign → ad group → ad) and a configured BI tool. Kondado delivers that data ready for the assistant, which is why it answers media questions no creative-spy tool can reach: the metrics are from your own account.

How do you connect TikTok Ads to ChatGPT and Claude?

The connection takes three steps, and none of them requires code.

1. Replicate your TikTok Ads into Via Kondado. You turn on the TikTok Ads data source in Kondado and Kondado materializes your account's report tables (with incremental replication by metric_date) and creative tables into Via Kondado.
2. Connect ChatGPT or Claude to the MCP endpoint. In Claude Desktop, in ChatGPT or in any MCP-compatible client, point to https://mcp.kondado.io/mcp and authorize access to your Via Kondado. Authentication is OAuth 2.1: the client registers through Dynamic Client Registration, or you paste a Bearer token. The endpoint exposes your TikTok Ads tables for the assistant to query, read-only.
3. Ask in plain language. From there, "which ads have passed frequency 3 and dropped CTR this week?" becomes a direct answer. The assistant writes the query, reads the tables and returns the number.

13 TikTok Ads analyses you run just by asking ChatGPT

With the tables in Via Kondado, the assistant answers questions that used to require exporting CSVs and building spreadsheets. Thirteen direct examples, each with the question you paste into the chat and the decision it unlocks:

• ROAS ranking per campaign. "Which are my 10 campaigns with the best ROAS in the last 30 days?" — the assistant reads complete_payment_roas (the ROAS the TikTok API already delivers) per campaign in tiktokads_reports_campaign_report, alongside spend, and sorts high to low. Decision: where to reallocate budget next week.
• Creative fatigue. "Which ads have passed frequency 3 and dropped CTR in the last 7 days?" — filters frequency > 3 in tiktokads_reports_ad_report, compares the window's ctr against the prior one and joins tiktokads_ad_details for the preview_url. Decision: the list of creatives to pause or rotate before CPM spikes.
• Video retention per creative. "Which 20 ads have the best retention to the end of the video?" — computes the video_views_p100 / video_views_p25 ratio per ad and brings preview_url and ad_format. Decision: which editing pattern (hook, pacing, CTA) to replicate in the next pieces.
• Full conversion funnel. "Show the funnel per campaign: view content, add to cart, checkout, purchase — and each stage's bottleneck." — aggregates view_content, add_to_wishlist, checkout, complete_payment and purchase per campaign. Decision: which stage to optimize (landing page, offer or pixel).
• Hourly performance (dayparting). "At what hour of the day did campaign X have the best CTR and the lowest cost per purchase?" — uses the hourly granularity (metric_datehour) and aggregates ctr and cost_per_purchase by hour. Decision: campaign scheduling adjustment.
• Creator and identity comparison. "Comparing the creators who post through my Spark Ads, which identity has the best average cost per purchase?" — joins report and details per ad and groups by display_name and identity_type. Decision: which creator to renew or scale.
• Budget pacing. "My campaign has BRL 30k for the month. How much have I spent and at what pace am I pacing?" — sums spend in the current month and projects the pace to period end, comparing with budget in tiktokads_campaign_details. Decision: speed up or slow down spend.
• Landing page view efficiency. "Which campaigns have the best landing_page_view_rate?" — sorts by landing_page_view_rate filtering total_landing_page_view > 0. Decision: spot landing-page load problems or a creative-to-offer disconnect.
• Auction vs reservation. "Comparing my auction campaigns with reservation buys, which model has the better CPM?" — runs a UNION ALL between tiktokads_reports_campaign_report and tiktokads_reports_reservation_campaign_report and compares average cpm. Decision: allocation between buying models, useful for awareness campaigns with guaranteed reach.
• Spend anomaly with no return. "Is any campaign spending more than BRL 500 per day with zero conversions in the last week?" — aggregates spend and complete_payment per campaign over the last 7 days and returns rows with high spend and zero conversion. Decision: pause the problem campaign before the weekend.
• UTM analysis. "Which content UTM values are bringing the most purchases?" — joins tiktokads_ad_details_utm_params with the ad report by ad_id and aggregates purchase per UTM value. Decision: consolidate the UTM naming convention that works.
• TikTok Ads vs Meta Ads comparison. "In the last week, which channel had the lower cost per purchase — TikTok Ads or Meta Ads?" — combines tiktokads_reports_campaign_report with facebook_campaign_insights (the Meta Ads table in the same Via Kondado) and computes cost per purchase per channel. Decision: where the marginal spend yields more conversion.
• Pixel vs real e-commerce. "Compare the complete_payment the TikTok pixel reported last week with the actual paid orders in Shopify." — confronts the report's complete_payment with shopify_orders filtered by paid status. Decision: detect the gap between what the pixel reports and the real order.

A sample question to paste into ChatGPT or Claude with your TikTok Ads connected:

From my TikTok Ads tables, show the 10 campaigns with the best ROAS in the last 30 days:
use the complete_payment_roas metric in tiktokads_reports_campaign_report,
alongside spend per campaign, and sort high to low.
Filter from D-2 onward to avoid still-incomplete data.

What to consider in your TikTok Ads data

Before drawing conclusions with AI, it's worth knowing a few platform behaviors — not as problems, but as good practices so your analyses stay reliable.

• The TikTok API has a 24-to-72-hour latency. Analyses about "today" will come with incomplete data. Ask the assistant to consider numbers from D-2 onward when you want firm conclusions.
• Spark Ads don't carry a preview_url. The video lives in the creator's account, not in the advertiser's media library — that's a limit of TikTok's own API. Retention and creator analyses work normally; only the visual link is empty in those cases.
• Landing page fields are recent. The total_landing_page_view, cost_per_landing_page_view and landing_page_view_rate metrics were added in February 2026; pipelines created before that may need to be recreated to see them.
• Paused and archived campaigns are replicated too. If you want to analyze only the active ones, ask the assistant to filter by secondary_status.
• The API delivers follows (the followers the ad generated), but not the growth of your total follower base over time. follows counts the new followers attributed to the ad; the history of your total follower base is organic data, outside the scope of paid media.

Official TikTok MCP vs Kondado MCP: what is the difference?

There are two MCPs in the TikTok world, and they solve different things. One manages campaigns; the other analyzes your already-replicated data.

The official TikTok MCP serves to operate the ads account. Kondado delivers the analysis: your media data — spend, ROAS, funnel, video retention and UTM — in Via Kondado, ready for ChatGPT and Claude to answer, without you opening a spreadsheet.

Anyone running TikTok Ads alongside other channels gains even more: with Meta Ads, Google Ads and Shopify in the same Via Kondado, the assistant compares cost per purchase across channels and validates the TikTok pixel against real paid orders — all in the same conversation. See the TikTok Ads analysis with AI for the full picture.

Frequently asked questions

Does TikTok have an official MCP server? Yes, TikTok has an official MCP server, but its scope is managing and operating campaigns inside the ads platform. To analyze your account's performance history — ROAS, funnel, video retention, UTM — the path is the Kondado MCP at https://mcp.kondado.io/mcp.

Can ChatGPT read my TikTok Ads campaigns? Yes, when your TikTok Ads account is replicated by Kondado into Via Kondado. ChatGPT does not talk to the TikTok API: it reads the tiktokads_reports_campaign_report, tiktokads_reports_ad_report and other tables, already materialized in Via Kondado, via the MCP endpoint.

Can I compute ROAS per campaign through ChatGPT? Yes. complete_payment_roas is a metric the TikTok API itself delivers in tiktokads_reports_campaign_report; the assistant reads that ROAS per campaign, factors in spend, and ranks by return — you just ask the question in plain language.

Can I compare TikTok Ads with Meta Ads in the same chat? Yes, when both channels are replicated in the same Via Kondado. The assistant combines tiktokads_reports_campaign_report and facebook_campaign_insights and computes the cost per purchase for each channel.

Does the data show up instantly? The TikTok Ads API has a 24-to-72-hour latency, so the current day's numbers come out incomplete. For reliable conclusions, ask the assistant to consider data from D-2 onward. Replication into Via Kondado runs at the frequency you choose.

Connecting your TikTok Ads to ChatGPT and Claude starts with one thing: having your campaign, creative and conversion tables in Via Kondado, where the assistant can query them in full. That is what Kondado delivers. Start your free Kondado trial and run your first TikTok Ads analysis in ChatGPT today.