Claude skill: build Dashboard Studio dashboards fully automatically
Hi all, and thanks @Satoer for this great app!
I’ve tested the new Dashboards Feature of Claude Claude Dashboards together with MCP AI Bridge and also Homey CLI Skill. It is really amazing of what Claude can do with this.
But data update is pull through the MCP Bridge only which is far from beeing efficient.
So I’ve tried if Claude now with its good knowledge of dashboards would be able to create Dashboards for the Dashboard Studio app, which is by far the most versatile Homey Dashboard system I’ve seen so far. But the Editor could be challenging for beginners (like me).
I’ve written a skill for Claude Desktop (Anthropic’s AI assistant) that designs, tests and saves complete Dashboard Studio dashboards on its own. You describe what you want (“show my Homey’s CPU, RAM and temperature with day and week graphs”), and Claude does the rest. It talks to Homey through the MCP AI Bridge app and works in the Dashboard Studio editor through its built-in browser.
@Satoer: I hope this is ok for you…
What the skill does
- Fully automated workflow: checks preconditions (MCP AI Bridge running, Dashboard Studio installed, editor URL from your Homey’s IP + port 5050), collects devices/capabilities/insights, generates the dashboard JSON, loads it into your editor with live values, checks it, and saves it to Homey after you confirm the name.
- Reads the real topic names from the editor’s live data / Data Stream Explorer. Names are never guessed from device names, so no more empty fields because of a wrong slug (umlauts, trailing spaces…).
- Enables the needed capabilities in Dashboard Studio automatically (
enabled_capabilities). It merges with your existing entries and never overwrites them. - Browser budget: on LAN addresses, Claude’s built-in browser asks you to approve every single browser action. The skill does everything it can via the bridge and bundles the browser work into ~4 calls: open editor, read topics, one test run with automatic checks (missing values, clipped text, formula errors), save + verify. So you click “allow” a handful of times instead of dozens.
- Data that isn’t a device capability (system values, insight history for 24 h / 7 days, min/avg/max) is fed in by an auto-created HomeyScript that pushes it as
homey/event/...topics. - Lots of verified Dashboard Studio know-how baked in: file schema, widget fields, formulas and bindings, graph quirks (
maxPointsdefaults to 20!), overlaying two graphs with a second scale on the right, gauges, navigation buttons, saving and verifying. - Safe by default: nothing is saved, overwritten, or created on Homey without your OK (unless your prompt explicitly allows it, see below).
Requirements / compatibility
- Claude Desktop with the integrated browser (that’s where I built and tested it), plus the MCP AI Bridge app connected to Claude.
- Dashboard Studio v2.0.12 on a Homey Pro (2026).
- I don’t know whether it works in other AI tools like ChatGPT. It relies on Claude’s skill format, the MCP AI Bridge tools and Claude’s built-in browser. Feedback welcome!
Example 1: Homey system monitor, built in one go
This prompt builds the dashboard in the first screenshot without any questions:
Use the dashboard-studio-json skill and the Homey MCP AI Bridge. Build a Dashboard Studio dashboard "Homey System" that monitors my Homey's internals, fully automatically and without asking me questions. You are explicitly allowed to: run test scripts, create one HomeyScript and one Advanced Flow as data feed, and save the dashboard to Homey under a new name (never overwrite an existing dashboard). Use the integrated Claude browser and stick to the skill's browser budget.
Content:
1. Current values as gauges: CPU load %, RAM used %, SoC temperature °C, flash storage used %; below each a short detail line (load 1/5/15 min, RAM used/total MB and swap, temperature range 24 h and 7-day max, free GB).
2. A power tile: undervoltage now (red, blinking) / OK (green), undervoltage max 24 h, undervoltage events since boot, throttling since boot.
3. Combined graph "CPU load & temperature" for 24 h (CPU as filled line) and for 7 days (CPU as hourly bars), temperature as an overlaid line with its own scale on the right; legend with avg/max CPU and temperature range.
4. Graphs of Homey's memory usage for 24 h and 7 days with avg/max.
5. RAM per app (top 5 table) plus Homey core and total apps memory.
6. Header with uptime and time of last update. Dark theme, 1280×870, scale layout.
Data: feed everything via a HomeyScript that sends values with the Dashboard Studio "Send to Dashboard Studio" card (homey/event/... topics): current values every minute, insight histories (system loadavg, temperature, memory, undervoltage ratio; 24 h and 7 days) plus min/avg/max every 5 minutes. Trigger it with an Advanced Flow every minute. Test everything in the editor, fix the layout until there are no missing values or clipped texts, then save, verify, and show me a screenshot.
What gets created on your Homey automatically:
- HomeyScript “DS SysMon”: reads
Homey.systeminfo (load average, RAM, swap, SoC temperature, uptime, undervoltage/throttling flags), memory per app, storage, and the system insights (CPU load, temperature, memory, undervoltage) for 24 h and 7 days. It pushes everything to Dashboard Studio ashomey/event/sys_*(current values),h_*(histories) ands_*(min/avg/max) topics. Current values update every run; histories every 5 minutes. - Advanced Flow “DS SysMon – every minute”: an every-minute trigger that runs the script. Disable it and the dashboard simply stops updating.
- The dashboard JSON, saved in Dashboard Studio (as a new file; nothing is overwritten).
Example 2: Energy flow dashboard
My energy dashboard was built with the same skill. It shows three Kostal inverters (with per-string power and daily yield), house consumption, the Tibber Pulse grid meter, the go-e wallbox and today’s totals (PV, consumption, grid import, cost).
- Animated energy-flow arrows: they only animate while power actually flows, line width scales with the power, and the grid arrow reverses automatically between export and import (based on the meter’s sign convention, which the skill checks first).
- High-contrast flow styling: a dimmed base line with a bright yellow moving pulse (red for grid import), so even small power flows are easy to see.
- Glowing icons for active sources.
- All values are live device capabilities. The skill enabled them in Dashboard Studio automatically and read the exact topic names from the editor; no extra scripts were needed for this one.
The skill file can be found here: github.com/tnsturm/DashboardStudioClaudeSkill. Install it as a skill in Claude, connect the MCP AI Bridge, and try it. Happy to hear how it works for you!

