Codex, Claude, Gemini, ChatGPT, Cursor, or Windsurf calls DataUnmess through MCP and publishes a reusable dashboard instead of leaving insights inside a chat thread.

3 steps

to connect

0 tokens

on reload

Live

workspace artifact

Claude Desktop

You

Assistant

DataUnmess | query_data -> build_chart x5

MCP

DataUnmess

Revenue Overview

Postgres | last 30 days

live

Revenue

125.4k

+12.5%

Conversion

3.24%

-0.5%

Avg Order

$68

+5.2%

Users

8.4k

+6.1%

Sales by Category

Revenue Trend

Traffic Sources

Stacked Revenue

waiting for prompt...

A complete data platform

Four surfaces.
One workspace.

Pipelines, flowcharts, data science, and dashboards - all driven by your AI of choice through MCP. No data team required to wire it up.

Data Pipelines

Sketch the ETL, then watch your AI fill it in. SQL transforms, joins, aggregations, derives - validated on a sample before it runs against your warehouse.

Flowcharts

Diagram pipelines, decision trees, architecture. Mermaid in, dash flows out. Editable by AI or hand, with auto-layout and per-node icons.

Data Science

Soon

Conversational analysis on your data - cohort splits, regressions, attribution. Returns charts, not text walls.

Dashboards

Live multi-chart layouts. KPIs, time series, distributions, comparisons - re-querying your real warehouse on every reload.

What chat artifacts can't do

Not another artifact in a chat.

Claude.ai and ChatGPT already draw charts - for a conversation, with fake data, locked to one vendor. DataUnmess gives your AI a real warehouse, a persistent team workspace, and artifacts it can re-query every time you open them. No tokens spent on reload.

No AI-provider API tokens. Any AI.

Create a DataUnmess MCP key, then paste our hosted MCP endpoint into Codex Desktop, Claude Desktop, ChatGPT, Cursor, Claude Code, or Windsurf. Use the flagship model your subscription already unlocks through that client.

Live data, not frozen demos

Connect Postgres, MySQL, Snowflake, ClickHouse, BigQuery, CSV, Excel, Google Sheets, or any REST API. Charts carry a query spec and re-run on every reload - zero AI tokens to open a saved dashboard.

Teams, not conversations

Workspaces with viewer / editor / owner roles. Dashboards and flows belong to the team - not to the chat that generated them. Invite, share, revoke.

Lineage & flow editor

Trace how a metric was built. Model data pipelines visually with interactive lineage graphs. Data-engineering-grade, not chat-grade.

Executable data pipelines

Ask the AI to clean a Google Sheet ("dedupe by email", "split full_name", "normalize phone numbers"). It compiles a DuckDB plan, previews the result, and writes a NEW spreadsheet - your original is never touched.

20+ chart types

Bar, line, donut, area, scatter, treemap, funnel, radar, sankey, heatmap, waterfall, gantt, ridgeline, and more.

Data science exploration

Aggregate, filter, and visualize through conversation. The AI writes the SQL; you see the charts.

AI edits everything

"Remove the dollar sign." "Sort descending." "Switch to ocean theme." Every UI setting is also reachable through natural language.

15+ themes

Carbon, Ocean, Sunset, Glow, Ember, and more - or build your own. Switch themes via chat or the Style panel.

Undo works on AI edits

Every AI action goes through the same history stack as a manual edit. Ctrl+Z rolls back anything.

How it works

First visit
to first useful artifact.

Login and create a key

Open the MCP key page, sign in, and create a workspace key. The key is free during beta and stays in your local AI client config.

> Open /account/mcp-server

Paste one setup prompt

Choose Codex, Claude, Gemini, ChatGPT, Cursor, or Windsurf. The copied prompt asks the agent to install DataUnmess MCP with a placeholder token.

> Follow /connect-mcp and add DataUnmess MCP

Ask what to build first

Start with company discovery, a CSV, or an existing warehouse. DataUnmess proposes the next useful dashboard and saves the result in your workspace.

> Use my site and docs to propose 3 dashboards

First asks

The first prompt should guide the work.

Once the MCP server is connected, your first ask can be small: catalog the company, inspect a file, or build one dashboard. DataUnmess turns that into persisted workspace artifacts.

See MCP onboarding

starter prompts

>Start my DataUnmess.ai MCP onboarding using my company site and docs

>Create a dashboard for this file: C:\tmp\sales.csv

>List my datasets and suggest the best first dashboard

>Query the orders table and chart the top 10 customers

>Clean my Q1 leads sheet - dedupe by email and split full_name

>Add KPI cards, a revenue trend, and breakdown by region

>Change card id:3 to a bar + line chart

>Add the month-over-month difference to the revenue chart

>Sort the bar chart descending

>Switch to the sunset theme

Two ways to deploy

Connect your AI and use it,
or hire our onboarding.

Free beta - self-serve

Try free: connect your AI

Create a free MCP key, paste one prompt into your AI client, then ask DataUnmess to catalog your data or build the first dashboard.

Create a DataUnmess MCP key and keep it out of chat
Use Codex, Claude, Gemini, ChatGPT, Cursor, or Windsurf
Team workspace with viewer / editor / owner roles
Full chart library, themes, lineage, flows

Try free

Talk to us

Hire our onboarding setup

We plan and execute the data integrations, then build the insights and artifacts for you. Turnkey for teams with data but no engineer to spare.

We plan the data integrations for your source systems
We execute the SQL pipelines, data models, KPI definitions
We build the insights - dashboards, flows, analyses - around your team's questions
Ongoing iteration + support

Get a quote