Four prompts. One live dashboard.
Most "AI dashboards" are screenshots of an answer — generated once, frozen the moment they're asked for, stale by morning. A dvt dashboard is different: a live, versioned spec that an agent writes, validates, and applies against your actual warehouse. This page is the whole onboarding. No sales call, no setup wizard — just the prompts, in order.
How it works — two surfaces
It's worth being precise about where dvt sits, because there are two separate things happening, not one.
Your warehouse
The agent explores your data with access you already have — a warehouse CLI, your dbt docs, whatever's on the machine. dvt is deliberately agnostic about this surface: it doesn't model your data or own your semantics.
dvt
The agent turns what it learned into a live, versioned dashboard spec over MCP — schema-validated, diffable, reviewable, and rendered faithfully every time someone opens it.
You bring the access. dvt owns the pixels.
Set up dvt
Connect the dvt MCP server to your agent. Two paths get you there — add the server directly, or install the Claude Code plugin, which wraps the same connection in a guided command.
https://mcp.dvt.dev/mcpclaude mcp add --transport http --scope user dvt https://mcp.dvt.dev/mcp \
--header "Authorization: Bearer YOUR_KEY"/plugin marketplace add getdvt/claude-dvt
/plugin install dvt@claude-dvt
/dvt:connectYour in-app API keys page has the exact snippet for every client.
"Connect to my dvt workspace: add the MCP server at https://mcp.dvt.dev/mcp using my dvt API key. Then verify it works — read dvt://skill/onboarding, list what you can see in my workspace, and give me a two-sentence orientation."
Then hand it the rest
Each prompt builds on the last. Paste them in order, in the same session — the agent carries context forward from one stage to the next.
Explore what’s in the box
The agent lists your folders and dashboards, opens the example dashboards and reads the docs written into them, and checks which skills are installed. You get a plain-language read on how the examples are built — and a first opinion on what to build differently for your data.
"Explore my dvt workspace: list the folders and dashboards, open the example dashboards and read their docs, and check what skills are available. Then tell me how the examples are built and what you'd build differently for us."
Bring your warehouse
The agent asks which warehouse you use, checks whether its CLI is installed and authenticated on your machine (snowsql, bq, psql, …), and — once it has read access — explores your schema starting with the most interesting tables. It comes back with three dashboard ideas grounded in what your data actually contains, not a generic template.
"Let's get you access to our data. Ask me which warehouse we use, then check whether its CLI is installed and authenticated here (snowsql, bq, psql, ...). Once you have read access, explore the schema — start with the most interesting tables — and pitch me three dashboards worth building."
Build the whole thing
This is the validate → preview → apply loop: the agent writes the SQL, designs a multi-page dvt spec, validates it against the schema, previews the render before touching anything live, and only then applies it. It finishes with a short written narrative and comments on anything surprising in the data — and walks you through the result once it's live.
"Build the best of those ideas end to end: write the SQL, design a visually distinctive multi-page dashboard in dvt, validate and preview the spec before applying, and finish with a short written narrative plus comments on anything surprising in the data. Walk me through it when it's live."
Not a screenshot. A live, versioned dashboard spec — validated before it was saved, previewed before it was applied, and current every time someone opens it, because dvt runs the SQL against your warehouse at view time, not once at generation time.
Change something next week and the agent edits the one panel that changed, validates it, previews it, and applies it. The rest of the spec is untouched.
Follow the build
dvt is in founding research. Leave your email and we'll reach out when there's something to see — no newsletters, no noise.