Introducing the Basedash MCP server

Today we’re launching the Basedash MCP server, a single remote endpoint that turns Basedash into a tool any AI client can call.

Connect Claude Code, Cursor, ChatGPT, Windsurf, or any other MCP-compatible client to one URL, run through an OAuth flow, and Basedash shows up as a tool the moment you’re done. Ask data questions in plain English and get answers with numbers, charts, and reasoning, without leaving the chat or editor you already work in.

Why we built it

How people use data has changed. A year ago the question was “how do I get to an answer faster inside my BI tool?” Today the question is “how do I get the answer inside the agent I’m already talking to?”

Engineers debugging in Cursor want feature-usage numbers without context-switching to a dashboard. Product managers in Claude want to compare cohort retention while drafting a spec. Operators in ChatGPT want last week’s revenue alongside the rest of their workflow. Each of those moments is an analytics question, and forcing a tab switch to answer it interrupts whatever they were doing.

The Basedash MCP server brings BI to where people already are instead of pulling them back into a BI app. The agent gets a data analyst on call, and the team doesn’t have to switch tools.

How the MCP server works

It takes one URL and one OAuth flow:

1. Add the server. Paste https://charts.basedash.com/api/public/mcp into your client as a remote streamable-HTTP MCP server.
2. Authenticate. The first call opens an OAuth window in your browser. Sign in with your Basedash account, the client gets a scoped token, and you’re done.
3. Ask anything. The client can now call Basedash from inside the chat or editor. Behind the scenes, every call hits the same AI data analyst that powers Basedash chat.

claude mcp add basedash --transport http https://charts.basedash.com/api/public/mcp

You don’t need API keys, pasted tokens, or a proxy.

Two tools, full coverage

We deliberately kept the surface small, with two tools that each cover a lot of ground.

ask_question is a back-and-forth with your data analyst. Pose a question in plain English (“trial-to-paid conversion by signup source, last 12 weeks, top 3 sources”) and get an answer. It handles quick lookups, deeper trends, comparisons, and strategic analysis with the same engine that runs Basedash chat. It generates and validates SQL behind the scenes, returns numbers, charts, and reasoning, and continues the same chat across follow-ups.

get_data_sources shows what’s available before you ask. It lists every connected database, warehouse, and SaaS source in your workspace so the client knows where to point a question: direct databases like Postgres and MySQL, warehouses like BigQuery and Snowflake, and 750+ SaaS apps via Fivetran or MCP connectors.

Same permissions, every client

Our most important design decision was that MCP exposes whatever an account can see inside Basedash and nothing more.

• Workspace permissions, enforced. Every tool call respects the same access controls as your Basedash workspace.
• OAuth-authenticated. Each client gets a scoped, per-user token. There are no API keys to share, store, or rotate.
• Source-level visibility. If your account can’t see a data source in Basedash, neither can the AI client connected to it.

That means you can roll MCP out to engineering, product, and ops without rebuilding governance from scratch. Agents inherit the same permission model your team already trusts in Basedash.

What teams are using it for

We’ve been running the MCP server internally and with a handful of customers for the last few weeks. A few patterns keep showing up:

• Engineers in Cursor. Check feature adoption or error rates against the production database before shipping a change, without leaving the editor.
• PMs in Claude. Compare retention cohorts and pull MRR snapshots inline while drafting a spec or a roadmap doc.
• Operators in ChatGPT. Ask “what happened in the business this week?” and get a sourced answer back in the same conversation they use for everything else.
• Founders in Claude Code. Run a one-shot “is anything unusual today?” before standup, in the same surface they already prompt against the codebase.
• Anyone in Windsurf or a custom client. Same URL, same answers, same governance.

In each case, analytics comes to whichever agent you’re already talking to.

Getting started

The Basedash MCP server is available today for all Basedash workspaces.

1. Sign up for Basedash (or log in)
2. Connect your data sources
3. Add https://charts.basedash.com/api/public/mcp as a remote MCP server in your client of choice
4. Complete the OAuth flow and start asking questions

For install steps specific to Claude Code, Cursor, ChatGPT, Windsurf, and other clients, see the MCP server feature page and the docs.

What’s next

The MCP server is the second half of how we think about Basedash and the agent ecosystem. MCP connectors let external apps act through Basedash’s agent, and the MCP server lets external agents read through Basedash’s analyst. Together, they make your data and the actions around it reachable from whichever client your team prefers.

With AI chat inside Basedash, Insights for proactive findings, Automations for scheduled workflows, and the Dashboard Agent for end-to-end dashboard generation, Basedash is now a complete AI-native BI platform that works inside our app or inside any AI tool you bring.

Try the Basedash MCP server today in the AI client your team already uses.