Install the connector
Add Dashies to Claude Code, Codex, Cursor, or Claude web and desktop, so your AI tool can publish dashboards and keep them refreshing.
Dashies is driven from the AI tool you already use. Installing the plugin is the
supported path: it bundles the dashies authoring skill and the publish service
together, and the skill is most of the product's behaviour.
A bare connector is not the same install
Adding only the MCP server gives your AI the publish tools and the service's own
brief instructions, but not the dashies authoring skill, which is where the
authoring behaviour is actually decided: how to design a cube, which
materialization to pick, and when a measure is not safe to sum.
Install the plugin wherever your tool supports one. Use the connector-only path below only for surfaces that cannot install a plugin.
Pick your tool
Run both commands in Claude Code. The first registers the Dashies marketplace, the second installs the plugin from it.
/plugin marketplace add https://dashies.xyz/marketplace.json
/plugin install dashies@dashiesdashies@dashies is <plugin>@<marketplace>. Both are named dashies, which is
correct and not a typo.
This bundles the dashies authoring skill and the publish MCP. Your first
publish opens the browser to sign in, and there is no token to paste.
Run both commands in your shell.
codex plugin marketplace add Dashies-ai/dashies-plugin
codex plugin add dashies@dashiesThen sign in over OAuth:
codex mcp login dashiesThis bundles the dashies authoring skill and the publish MCP.
Cursor takes two steps, because it installs the connector and the skill by different routes.
First, add the publish connector with the one-click install link:
Then add the authoring skill:
npx skills add Dashies-ai/dashies-plugin --global --agent cursor--global installs at user level rather than into whatever directory you happen
to be in. It writes ~/.agents/skills/dashies/SKILL.md and
~/.agents/skills/dashies-design/SKILL.md, which is one of Cursor's user-level
skill roots. It works whether or not the cursor shell command is on your PATH.
Restart Cursor once both are done. Your first publish opens the browser to sign in, and there is no token to paste.
If the one-click link does not land, add the connector by hand to
~/.cursor/mcp.json:
{
"mcpServers": {
"dashies": { "url": "https://mcp.dashies.xyz/mcp" }
}
}Claude web, desktop, and Cowork add Dashies as a custom connector. A connector
can only add the MCP server, so on these surfaces the dashies authoring skill
is not installed. Publishing works; the authoring guidance that makes a
dashboard correct does not come with it.
The connector URL is:
https://mcp.dashies.xyz/mcp- In Claude, open Settings, then Connectors.
- Click the plus button, then Add custom connector.
- Paste the URL above, then click Add.
Sign-in opens once, and there is no token to paste.
One command auto-detects every plugin-capable AI agent on your machine and installs the Dashies plugin into each:
npx plugins add Dashies-ai/dashies-pluginThis always delivers the authoring skill, never a bare connector.
Check it worked
Ask your AI tool to list your dashboards. In plain language:
List my Dashies dashboards.
Three things should happen, in order:
- Your tool calls the
list_dashboardstool. - A browser window opens once, asking you to sign in with Google. See Sign in and authorize.
- You get back a list, which is empty on a new account. An empty list is a pass. The tool call completing at all is the signal.
If your tool reports that it has no Dashies tools, the plugin did not install. Re-run the install commands for your tab above and restart the tool.
A newly installed tool argument can be missing
An MCP client reads each tool's schema once, when the session connects. If you installed or upgraded Dashies inside a session that was already running, a newly shipped argument is simply absent, as if it had never been built. Start a new top-level session after installing.
Next
- Sign in and authorize if the browser flow did not complete.
- Connect a warehouse to make dashboards refresh from your own data.