A common query contract.
Platform-specific connectors sit behind shared discovery and query tools. The assistant can learn which metrics a source supports before asking it for data.
From the lab
Opening the build…
Ask the question. Reach the actual data.
Analytics integrationAI automation
A connection between AI assistants and the marketing platforms an agency works across.
Inside the build
The brief
The work behind the interface.
The situation
An agency’s answers often live across analytics, search, advertising, commerce, and CRM platforms. A useful question can turn into several logins and exports before anyone can explain what happened. Arlo gives the assistant a structured way to reach those systems.
What we built
We built an MCP platform with account connections, a connector registry, common query tools, and workspace context. The assistant resolves the client and reporting period, queries the connected source, and works from the returned records.
Inside the build
Follow the work.
01 / Connect
Bring the sources into the workspace.
Connectors cover platforms such as GA4, Search Console, Shopify, and HubSpot. Each connector defines the metrics, dimensions, and query behavior the assistant can use.
Open full capture ↗Inside the build
Behind the build
The decisions that shaped it.
Keep account boundaries explicit.
Workspace and caller context are part of the request path. Resolving the right business is a prerequisite to a useful answer.
Discover before querying.
Connector discovery exposes the metrics and dimensions a platform supports. The question can be grounded in the available data before the query runs.
What the build makes possible
A capability the business can use.
Next.jsTypeScriptMCPOAuthPlatform connectorsConvex
Arlo connects natural-language questions to structured queries against business platforms. This case study documents the implemented product architecture and workflow; it does not claim a measured reduction in reporting hours.
Answers depend on the connected account, available source data, and the question asked. Source context and tool visibility support review; they do not make every AI interpretation correct.