Welcome to Lighty

Explore your workflows, read the analysis, or drive Lighty from Claude.

🧭

Studio

Upload a dataset, discover workflows and causal structure, and explore the analysis interactively.

Open Studio →

📊

Reports

Browse the customer analysis reports — findings, charts, and recommendations, by engagement.

Open Reports →

🔌

Connect Claude

Drive Lighty from Claude — Desktop or Code — with the MCP server + skills, analyzing datasets in plain language.

Setup instructions →

Connect Claude — MCP + Skills

Install the lighty-dogfood plugin to drive the Lighty platform from your own Claude — Claude Desktop or Claude Code: point it at a dataset and get back discovered workflows, causal structure, and a report.

1 Install the plugin (MCP + skills)

The plugin bundles the MCP server connection and the analyze-dataset skill. It's published in a public marketplace, so there's nothing to download. Pick your client:

It all happens in the app — no terminal needed.

⚠️ The /plugin commands don't work in the Desktop chat — those are Claude Code only. In Desktop you install from the Plugins menu, below.

🧩 Add the marketplace · one time

Open the Customize menu in the left sidebar, then open the Plugins tab. (In Cowork, open the Cowork tab first, then Customize.)

In the Personal plugins section, click the small + button, then Create plugin → Add marketplace — yes, "Add marketplace" is nested under Create plugin:

Claude Desktop Personal plugins menu — the + opens Browse plugins and Create plugin; Create plugin expands to Add marketplace, Upload plugin, and Create with Claude — The small + next to **Personal plugins** → **Create plugin** → **Add marketplace**.

Then paste this and confirm:

LightyAI/lighty-plugins

The field accepts a GitHub owner/repo or git URL — the full link https://github.com/LightyAI/lighty-plugins works too.

🖱️ Install the plugin

Click the + again → Browse plugins to open the directory. Go to the Code tab, pick the lighty-plugins marketplace, and click Lighty dogfood → Install.

Claude plugin directory open to the Code tab and lighty-plugins marketplace, showing the Lighty dogfood plugin card by Lighty AI — Plugin directory → **Code** → **lighty-plugins** → **Lighty dogfood** (by Lighty AI). Click it to install.

🔌 Connect & sign in

In the sidebar, open Lighty dogfood → Connectors, find lighty-platform, and click Connect. That opens the WorkOS sign-in in your browser (Step 2) — sign in with your Lighty account and approve. Then restart Claude Desktop so the Lighty tools load.

Sidebar showing the installed Lighty dogfood plugin with Skills and Connectors entries — **Connectors** lives in the sidebar under the installed **Lighty dogfood** plugin.

The lighty-platform connector connected, showing its app.lighty.ai/mcp URL and a list of available tools — Connected looks like this — **lighty-platform** (`app.lighty.ai/mcp`) with its tools listed. Once you see **Disconnect** and the tool list, you're done.

Installing the plugin alone doesn't connect the server — skipping Connect is why lighty-platform won't show up in Claude's tool list.

More detail: the official Use plugins in Claude guide.

⌨️ In Claude Code — type these straight to Claude

These are Claude Code slash commands. Type them into the Claude Code message box, one at a time (they start with a /):

/plugin marketplace add LightyAI/lighty-plugins

/plugin install lighty-dogfood@lighty-plugins

Then type /reload-plugins (or restart Claude Code) so the MCP server connects.

⚠️ /plugin works only in Claude Code — paste it into Claude Desktop or claude.ai and you'll just get back "that's not a command I have." On Claude Desktop, use the Claude Desktop tab above instead. New to Claude Code? Install it with npm install -g @anthropic-ai/claude-code.

2 Sign in with WorkOS

The first time Claude connects, it opens a WorkOS sign-in in your browser — sign in with your Lighty account and approve. No tokens to copy.

You need to be a member of the Lighty organization (the same one you use for this dashboard). Ask Anthony if you're not sure.

3 Use it — then watch it in the Studio

Ask Claude in plain language — no flags or tool names needed:

"Run the sample dataset through Lighty and tell me what you find."
"Show me the datasets Lighty has already analyzed."
"Analyze the support-tickets dataset and tell me what drives slow resolutions."

No data of your own? The first line runs a bundled sample end-to-end — zero setup. Every analysis you launch shows up live in the Studio: it appears in the Live rail with streaming phases as it runs, and links straight to the finished report.

v0 dogfood — pipeline-grade results. Bring your own file via the Explorer (drag-and-drop), then tell Claude the dataset name. Found a rough edge? That's the feedback we want.