Asemic exposes an MCP server (Model Context Protocol, the open standard AI assistants use to call external tools). Connect it once, and you can ask analytics questions in the AI chat you already use:

“What was ARPDAU by payment segment last week, and how does it compare to the week before?”

The AI discovers your project’s metrics and dimensions, runs the query through Asemic, and narrates the result. It can also find and read existing dashboards and stories, and hand you a link to open them in Asemic.

The reason the answers are trustworthy: the AI does not write SQL against raw tables. It queries the named, versioned metric definitions in your semantic layer, so the numbers in chat match the numbers on your dashboards, whoever asks.

Works with claude.ai (custom connectors), Claude Desktop, Claude Code, and any MCP-capable client.

Current scope is read and query. Creating and editing dashboards from chat, and receiving charts as images, are on the roadmap.

Connect from claude.ai or Claude Desktop

In claude.ai, open Settings, then Connectors, then select “Add custom connector”.
Paste the server URL: https://app.asemicanalytics.com/mcp
The browser opens Asemic’s sign-in (your regular Asemic credentials or Google sign-in), followed by a permission screen listing what the connection may do:
- “Discover metrics and dashboards, run analytics queries, render charts”
- “Create and edit dashboards and stories”
Select “Authorize”. The chat can now use the Asemic tools.

A few things worth knowing:

No API key is created or pasted anywhere in this flow; registration and sign-in are automatic (OAuth).
Access is per user: the AI sees exactly the projects you can see, nothing more.
The second permission (“Create and edit dashboards and stories”) is granted now, but the write tools ship later; today the integration reads and queries only.

Connect from Claude Code

Power users connect with a personal access token instead of the OAuth flow:

In Asemic, create a token under the profile settings, “Personal Access Tokens”.
Add the server:

claude mcp add --transport http asemic https://app.asemicanalytics.com/mcp \
  --header "Authorization: Bearer <your-token>"

Tokens are revocable from the same settings page.

What you can ask

“What metrics do we track?”
“Show DAU for the last 30 days, split by platform.”
“Compare ARPDAU this month vs last month for payers only.”
“What does our retention curve look like for the June cohorts?”
“Do we have a dashboard about monetization? Link me to it.”

Tool reference

These names appear in your chat client’s tool list:

Tool	What it does
`list-projects`	Lists the projects you can query.
`list-metrics`	Lists a project’s metrics with descriptions and units, and whether each plots over calendar dates, days since install (cohort), or both.
`list-dimensions`	Lists the user properties available for grouping and filtering.
`get-dimension-values`	Lists the distinct values of one dimension (for exact filters).
`run-query`	Runs a query: one or more metrics over a date range, optionally grouped by dimensions, filtered, and compared against the previous period. Returns a compact table plus totals.
`list-dashboards`	Lists dashboards and stories in a project, with links.
`get-dashboard`	Reads one dashboard or story: its charts, settings, and story text.

How the AI behaves

Results are capped (default 200 rows). The AI is told when a result was truncated and how to narrow the query, so you may see it refine its own query; that is expected.
Daily metrics plot over calendar dates; cohort metrics plot over days since install. The AI is told which is which, so questions like the retention example work naturally.
If a query is invalid (unknown metric, bad filter), the AI receives a precise error and corrects itself. This is normal and visible in the chat’s tool log.

Security and privacy

Sign-in uses your existing Asemic account, and the permission screen is shown before any access.
Query access is read-only. The AI cannot change the data model, connections, or settings through this integration.
Queries run on demand against the project’s published model; raw event data is not copied to the AI provider beyond the query results shown in chat.
Access tokens expire and refresh automatically (OAuth), or are revocable personal tokens on the Claude Code path.