TL;DR
Mitzu's MCP server exposes the Analytics Agent as a backend for any MCP-compatible agent — Claude, Cursor, ChatGPT, or a custom agent. Mitzu's own Analytics Agent runs the full investigation end-to-end. No tool orchestration on the calling agent's side. The deterministic SQL engine generates the query underneath — the answer is grounded in product analytics methodology, not inferred by an LLM.
Mitzu now answers product analytics questions from inside whichever AI agent you already use. Open Claude, Cursor, ChatGPT, or the agent your team built — and Mitzu's Analytics Agent is right there in the chat. Funnels, retention, segmentations, journeys: the same analyses you'd run inside the app, surfaced wherever the work is already happening.
The MCP server is live. This post covers what it does, how it works under the hood, and who it's for.
1. What Mitzu's MCP server does
Most MCP servers give an agent a toolbox to dig through — raw tools the agent has to discover, select, and chain together to get to an answer. Mitzu's MCP server gives the agent an analyst.
When a question lands, Mitzu's own Analytics Agent runs the full investigation end-to-end. No tool orchestration on Claude or Cursor's side, no skill chains to write. The question goes in; a grounded answer comes back.
The analyses available are the same ones available in the app: funnels, retention curves, segmentations, user journeys, cohort comparisons, root cause investigations. The same semantic layer, the same deterministic query engine, the same methodology — just surfaced from inside the agent you're already using.
2. How it works
The trust differentiator in Mitzu has always been the deterministic query engine. The Analytics Agent doesn't write SQL — it assembles analysis specifications (funnel steps, retention windows, segmentation breakdowns) and hands them to an engine that has been generating methodologically correct SQL for years. Same specification, same SQL, same answer, every time.
That mechanic is exactly what runs behind the MCP server. When Claude or Cursor forwards a question to Mitzu, Mitzu's Analytics Agent picks it up, runs the investigation using the full product analytics methodology, and returns the result. The calling agent doesn't need to understand funnels, cohort time-bucketing, or conversion windows — Mitzu's engine handles that.
The answer is grounded in product analytics methodology, not inferred by the calling LLM. A funnel with a correct conversion window, a retention chart that enforces cohort time bucketing, a segmentation that doesn't double-count users — these aren't things Claude or Cursor need to get right. Mitzu's engine gets them right.
3. Durable conversations and first-class artifacts
Every exchange through the MCP server also becomes a real Mitzu conversation behind the scenes — durable, with a shareable URL your team can open, continue, or hand off. The conversation doesn't disappear when the chat session ends.
The underlying insights the agent built along the way — the funnels, the segmentations, the journeys — live on as first-class Mitzu artifacts. Visit them in the app, save them to a dashboard, or return to them later. Nothing gets lost between the AI agent and the analytics workspace.
4. Who it's for
Because anyone can ask from inside the agent they already use, Mitzu's MCP server is useful across the whole organisation — not just the people who open the analytics app every day.
- Customer success — checking on at-risk accounts while prepping a call, without switching tabs.
- Marketing — pulling live campaign metrics into a planning doc, from inside the tool where the doc is being written.
- Product managers — grabbing post-release usage stats for a deck, directly from the IDE or chat.
- Engineers — prioritising bug fixes using product data, right from the IDE where the code already lives.
- Data analysts — routing questions from team members who prefer Claude or Cursor to learn-a-new-UI, while keeping all the results in Mitzu's workspace.
This is the same pattern as Mitzu's Slack Agent — meeting people where they already are rather than asking them to open another application. The MCP server extends that to any agent that speaks the Model Context Protocol.
5. Compatible agents
Mitzu's MCP server works with any MCP-compatible agent: Claude (Desktop and claude.ai), Cursor, ChatGPT, or a custom agent your team has built. Setup instructions are in the Mitzu MCP documentation.
6. Warehouse readiness — the prerequisite
The MCP server runs on the same foundation as the rest of Mitzu: event data in a modern cloud warehouse — Snowflake, BigQuery, Databricks, Redshift, or ClickHouse — with Mitzu's Configuration Agent having indexed it into the semantic layer. No data leaves the warehouse for analysis to happen. If event data still sits in a third-party vendor silo without warehouse export, Mitzu isn't the right fit yet.
Frequently asked questions
What is Mitzu's MCP server?
Mitzu's MCP server exposes Mitzu's Analytics Agent as a backend for any MCP-compatible agent — Claude, Cursor, ChatGPT, or a custom agent. When a product analytics question arrives via the MCP server, Mitzu's own Analytics Agent runs the full investigation end-to-end and returns a grounded answer. The calling agent doesn't need to orchestrate tools or understand product analytics methodology.
Does the calling agent (Claude, Cursor) write the SQL?
No. Neither the calling agent nor Mitzu's Analytics Agent writes SQL. The Analytics Agent assembles an analysis specification — funnel steps, retention parameters, segmentation breakdowns — and Mitzu's deterministic query engine turns that into SQL using product analytics methodology developed over years. The same specification always produces the same SQL.
The calling agent just asks the question.
Are conversations and insights saved?
Yes. Every exchange through the MCP server creates a real Mitzu conversation — durable, with a shareable URL. The underlying insights (funnels, retention charts, segmentations, journeys) built during the conversation are saved as first-class Mitzu artifacts. They can be added to dashboards, shared with the team, or picked up later in the app.
Which agents and tools are compatible?
Any agent that implements the Model Context Protocol: Claude Desktop, claude.ai, Cursor, ChatGPT, and custom agents. See the Mitzu MCP documentation for setup instructions per agent.
Does this require a specific Mitzu plan?
The MCP server is available for workspaces with the Analytics Agent enabled. See Mitzu pricing for plan details.
Get started
Setup takes a few minutes. Connect your agent to Mitzu's MCP server, point it at your workspace, and start asking questions. Full instructions are in the Mitzu MCP documentation.
Related reading
- Planning Mode in Mitzu — review the agent's plan before it runs
- Custom Charts — ad hoc visualizations inside the Analytics Agent
- Why agentic analytics needs a product-analytics-shaped semantic layer
- What is agentic analytics?
- How AI agents query your data warehouse
