Lightdash agent plugin

This plugin connects AI coding agents to the Lightdash MCP server and adds an analytics workflow skill. It is intended for governed metric discovery, semantic-layer queries, and creating analytics content safely.

Source: github.com/lightdash/claude-plugin. The plugin lives at the repository root. One bundle serves three marketplaces:

• Claude (Claude plugin directory and Claude Code): .claude-plugin/plugin.json
  • .mcp.json, discovered via .claude-plugin/marketplace.json at the repository root
• Codex (OpenAI plugins directory): .codex-plugin/plugin.json + codex.mcp.json
• Cursor (Cursor marketplace): .cursor-plugin/plugin.json + mcp.json, discovered via .cursor-plugin/marketplace.json at the repository root

Shared between both: skills/lightdash-analytics (guides the agent through discovery, querying, and explicit confirmation before workflows that change analytics content) and assets/ (brand images).

Install in Claude Code

/plugin marketplace add lightdash/claude-plugin
/plugin install lightdash@lightdash

Claude Code asks for your MCP server URL when the plugin is enabled. Paste the full URL of your instance, Cloud or self-hosted:

https://<workspace>.lightdash.cloud/api/v1/mcp
https://<your-lightdash-host>/api/v1/mcp

Then run /mcp and complete the Lightdash sign-in. Non-interactive installs pass the value directly:

claude plugin marketplace add lightdash/claude-plugin
claude plugin install lightdash@lightdash --config mcp_url=https://<workspace>.lightdash.cloud/api/v1/mcp

In claude.ai, Claude Desktop and Cowork the per-workspace URL comes from the Lightdash MCP connector in the Claude directory, not from this plugin: the connector listing uses a URL pattern, so each user enters their own instance URL and signs in with OAuth (client ID metadata documents, with dynamic client registration as the fallback).

Authentication and instance URL

The MCP endpoint uses OAuth. Install the plugin and complete the Lightdash sign-in prompt in your agent.

Cursor: mcp.json reads the endpoint from the LIGHTDASH_MCP_URL environment variable, so one config serves every Lightdash instance — Cloud workspaces and self-hosted alike. Before first use, set it to your instance:

export LIGHTDASH_MCP_URL="https://<your-workspace>.lightdash.cloud/api/v1/mcp"
# self-hosted: https://<your-lightdash-host>/api/v1/mcp

Codex: codex.mcp.json targets Lightdash Cloud (app.lightdash.cloud). For a different workspace or a self-hosted instance, replace the URL with https://<your-lightdash-host>/api/v1/mcp — the Codex manifest format has no URL templating today.

What the plugin does

The plugin contains no code that runs on your machine. It declares one remote MCP server (the URL you enter) and one skill, a Markdown file that guides the agent. All requests go to your Lightdash instance over HTTPS after you sign in with OAuth; nothing is sent anywhere else.

License

MIT. See LICENSE. Copyright Telescope Technology Limited (trading as "Lightdash").