MCP: your data inside an AI assistant
Traffic Insights now runs a Model Context Protocol server. Once an assistant such as Claude or ChatGPT is connected to it, you can ask about a project in chat and the assistant pulls the real numbers from your workspace instead of guessing: project list, project details, the latest analysis with scores and anomalies, saved reports and connected sources.
Access is authenticated: the assistant signs in with your account through OAuth, and every query is filtered by row-level security so it can only ever read your own data. All five tools are read-only, so an assistant can analyse and summarise but cannot change or delete anything, and integration credentials are never exposed.
- list_projects — all projects in your workspace
- get_project — one project with its metrics
- get_analysis — the latest analysis, scores, segments and anomalies
- list_reports — saved reports
- list_integrations — connected sources and their sync status
Private Google Sheets without making them public
Sharing a spreadsheet by public link was the only option before. Now you can connect your Google account once in Settings and read spreadsheets you already have access to, keeping them private. The connection is stored encrypted per user, can be disconnected at any time, and the public-link export still works as a fallback for teams that prefer it.
Scoring beyond GEO: affiliates, sources, campaigns, offers
The deterministic scoring engine no longer assumes country is the only useful dimension. It inspects the dataset, recognises columns describing affiliates, partners, buyers, sources, campaigns and offers, and scores each value from 0 to 100 using ROI, FTD economics, spend, revenue, average check and share of volume. Every dimension gets its own leaderboard with a Scale, Test or Cut recommendation, so a profitable GEO hiding two loss-making buyers becomes visible immediately.
Clearer red flags and a guide inside the product
Anomalies used to be reported in technical shorthand. Each flag now says which affiliate, GEO or campaign it concerns, what exactly looks wrong — spend without conversions, payback collapse, duplicated rows, a statistical outlier — and what to verify before acting on it.
A guide is now built into the interface. It walks through uploading data, connecting sources, running a KPI check, reading scores and anomalies, using the assistant and connecting an AI assistant over MCP, and links to the public expert guides. Privacy Policy and Terms of Service pages are published as well.