Preset puts conversational analytics on Apache Superset with Claude Code

"Data analytics is changing really quickly. Claude has gotten really good at analyzing data, visualizing it, giving you insights into it. That's why for us, building an MCP on top of Superset, building a chatbot within the product, there's so many doors opening up." — Elizabeth Thompson, Director of Engineering, Preset

Challenge

Customers want plain-language answers on top of Apache Superset

Preset sells enterprise analytics on Apache Superset: customers bring their own data, their own semantic layer, and as many as 50 to 75 databases, then build visualizations on top. The BI category is moving off that dashboard. Instead of a human staring at a chart and inferring the next decision, customers expect to ask questions in plain language and, next, to have the product suggest or take the action.

That surface has to be built on the same open-source codebase Preset contributes back to. Preset's  engineering org is comprised of three teams building their enterprise analytics product. Two work on Superset, and the third is a platform DevOps team that also works on Superset. Each team is two to four people. Years of accumulated contributors make the code hard to keep clean.

Solution

Remotely engineers helped build the chatbot and the Superset MCP in Claude Code

Remotely.Works embedded engineers across product engineering, QA, and DevOps/security.

An MCP on Apache Superset is the way a customer asks questions of their own data, semantic layer, and visualizations. Preset contributed that MCP back to Apache Superset, so Claude and other MCP clients can use it on the open-source project, not only inside Preset. A chatbot inside Preset uses the same MCP and integrates OpenRouter, so the customer can also ask from inside the product, on the model they choose. Engineers moved from IDE plugins to the terminal and Claude Code, and Claude Code is how this surface gets built.

"Our product engineers are working on a chatbot, so they're not only using AI, but they're building AI features."— Elizabeth Thompson

The governance layer for HIPAA customers has since been delivered.

The same small teams keep that codebase shipping. A Remotely QA engineer runs a bot that spins up a local git worktree, drives Playwright, and takes screenshots to confirm a fix. Remotely DevOps and security engineers remediate vulnerabilities the security tooling surfaces, and keep up with patching.

"We have Remotely team folks who are helping to remediate those vulnerabilities, as well as just keeping up with all of the patching that needs to be done. On the QA side, we have a really great bot that our engineer is using to spin up a local work tree, that can use Playwright to then take screenshots to validate if something is working or if a bug has gotten fixed."— Elizabeth Thompson

How it works

1. An enterprise customer brings their own data, semantic layer, and database into Preset on Apache Superset.
2. An MCP on Apache Superset exposes that analytics surface. Preset contributed the MCP back to the open-source project, so an MCP client such as Claude can query it.
3. A chatbot inside Preset uses the MCP and integrates OpenRouter, so the customer can ask questions in plain language on the model they choose.
4. A governance layer for HIPAA customers has been delivered on that stack.

A Remotely QA engineer confirms fixes with a local worktree and Playwright screenshots. Remotely DevOps and security engineers remediate vulnerabilities on the same codebase.

Impact

Customers can now ask questions of their own data through Preset MCP, including from Claude, and through a chatbot inside Preset.

Preset MCP handled about 100,000 customer tool calls in Q2 2026 (May through July). Preset's own use is excluded. 87% of those calls came from Enterprise customers.

32% of Enterprise customers have been using the MCP in the last 90 days.

Using Claude Code, the rollback rate fell by half.

‍