[← All posts](/blog)
If you’re a SaaS product team, you’ve almost certainly evaluated embedded analytics, the category of tools that let you drop dashboards and reports into your product so customers can see their data. It’s a mature, well-understood decision.
A newer option is now on the table: exposing your data to your customers’ AI agents through an analytics MCP server. It sounds adjacent, and the two often get lumped together, but they solve different jobs. This piece lays out the difference so you can tell which one your product needs, and when the answer is both.
The core distinction #
Embedded analytics ships a dashboard to a human. The output is a visual experience (charts, filters, a report builder) rendered inside your product. Your customer looks at it and clicks.
MCP analytics ships governed data access to an agent. The output is structured answers to questions your customer’s AI agent asks, through the Model Context Protocol. No chart is rendered; the agent consumes the data and reasons over it.
Same underlying metrics, in many cases. Completely different consumption surface.
A quick way to tell them apart #
Ask who is on the other end of the request, and what they want back:
| Embedded analytics | MCP analytics | |
|---|---|---|
| Consumer | A human | An AI agent (often the customer’s own) |
| Output | Rendered charts and dashboards | Structured, governed answers |
| Interaction | Point, click, filter | Programmatic queries via MCP |
| Primary job | ”Show me my data" | "Let my agent work with your data” |
| You control | The full visual UX | The metrics, permissions, and query surface |
Where each one wins #
Reach for embedded analytics when your customers want to see their data inside your product: operational dashboards, reporting tabs, KPIs at a glance, a self-serve report builder. The value is a polished, native visual experience, and the person consuming it is a human who benefits from good design.
Reach for MCP analytics when your customers want their agents to work with the data your product holds: pulling their own records into an automated workflow, letting an in-product copilot answer live questions, or connecting your product to the customer’s broader agent stack. The value is programmatic, governed access, and the consumer is software.
The tell that you need MCP analytics specifically: customers start asking whether their AI tools can connect to your product, not just whether they can view their data in it.
Why “just expose an API” isn’t the same answer #
A reasonable objection: don’t we already have a REST API for programmatic access? You might. But an analytics MCP server differs in ways that matter for agent access:
Discovery. MCP lets an agent discover what metrics and dimensions are available in a standard way, rather than requiring the customer to read your API docs and hand-code an integration.Governed metrics. A raw API often exposes tables or endpoints an agent can misinterpret. An analytics MCP server exposesdefinedmetrics, so answers stay consistent with what your product’s own UI reports.Agent-shaped safety. Multi-tenant isolation, identity propagation, rate limiting, and audit logging designed for autonomous callers that query far more aggressively than a human integrator would.
So it’s less “we already have an API” and more “an analytics MCP server is the governed, agent-ready front door to the same data.”
In practice, many products will want both #
These aren’t mutually exclusive, and the strongest setups share a foundation. If you define your metrics once, in a governed semantic layer, you can serve them to an embedded dashboard and to a customer-facing MCP server, and both will report the same numbers. That shared source of truth is what keeps a human looking at a chart and an agent querying an endpoint from ever disagreeing.
So the honest answer to “which do I need?” is often sequencing, not exclusivity: embedded analytics if the immediate demand is visual reporting; an analytics MCP server as soon as customers start bringing agents. Products serving both audiences will run both.
Where Motley fits #
Motley builds the shared foundation and the agent-facing surface. SLayer, our open-source semantic layer, is where you define your metrics once so every consumer sees the same governed numbers. Motley is the hosted platform on top that runs the multi-tenant MCP endpoint your customers’ agents call, with per-tenant scoping and audit built in. If you already run embedded dashboards, Motley is the agent-facing complement on the same metric definitions; if you’re starting fresh and your customers are agent-first, it can be where you begin.
Wondering where a managed MCP server fits alongside, or ahead of, your embed? Book a demo.