Connect Claude to your product analytics with MCP

Most analytics tools force you to build dashboards before you can answer a question. With MCP, you can skip the dashboard step entirely — just ask Claude. Here’s how to set it up against your Inslytic project in about three minutes.

What MCP is

MCP (Model Context Protocol) is Anthropic’s open standard, released in late 2024, for giving AI assistants structured access to external tools and data. An MCP server exposes a small set of typed tools; an MCP client (Claude Desktop, Claude Code, Cursor, and a growing list of others) discovers those tools and calls them on the user’s behalf. The protocol is transport-agnostic — it works over stdio for local processes and over HTTP for remote services — and the tool surface is intentionally tiny: a name, a JSON schema, a description.

Inslytic ships an MCP server with two tools. get_event_taxonomy returns your product description, your configured events, and the names of every event actually present in ClickHouse, so the assistant queries against real fields, not hallucinated ones. query_events runs read-only SQL against your events table, scoped to your project. The server injects your project ID and date range into the query — the assistant never sees another tenant’s data, and it can’t mutate yours.

Why MCP changes product analytics

Traditionally, answering “how is activation trending?” means a PM either writes SQL, clicks through a funnel builder, or pings the data person. Each path adds latency between question and answer, and most questions die in the queue. MCP flips the model: the assistant reads the schema, writes the query, runs it, and explains the result — in the same chat where you’re already working. The dashboard becomes a fallback for visual exploration, not the primary surface. Most product questions never need a chart at all; they need an answer and a next step.

The other thing MCP changes is who can ask. A founder who can’t write SQL, a designer debugging an onboarding step, a support lead spotting a regional issue — they all get the same query interface as whoever owns the data. The bar to checking the data drops from “know SQL” to “know what you’re curious about.”

What you can ask

Here are three prompts that work the day you wire MCP up. The replies below are illustrative, but the shape is real — Claude reads your event taxonomy, writes a SQL query against ClickHouse, and explains the result.

Setup in 3 steps

1. Sign up at inslytic.com — the free tier covers up to 10K events/month, no credit card needed.
2. Grab your MCP URL and project API key from Settings → Integrations in the Inslytic dashboard. Both are scoped to a single project.
3. Add the server to your MCP client. For Claude Desktop, open claude_desktop_config.json and add this block:

{
  "mcpServers": {
    "inslytic": {
      "url": "https://api.inslytic.com/v1/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_PROJECT_KEY"
      }
    }
  }
}

Restart Claude Desktop, and Inslytic will show up as a connected tool. For Cursor, Claude Code, and the full list of clients (including the mcp-remote shim for stdio-only clients), see the MCP server page.

What else this unlocks

Once an MCP server is wired up, “ask Claude” becomes a primitive your team composes with. A few patterns that show up early:

• Weekly product summaries posted to Slack by an agent that reads MCP each Monday morning and flags the deltas that matter.
• Post-deploy impact analysis without setting up dashboards for every release — ask after each merge, get a before/after read.
• Onboarding cohort comparisons on demand — “is the v2 flow doing better than v1 for users who signed up via the AI landing page?”
• Exec-team questions answered in the chat thread, without pinging the data person at 11pm before the board meeting.

Get started

Inslytic is AI-first product analytics with a native MCP server. The free tier is generous, the infrastructure is EU-hosted, and there’s no credit card required to try it.