Create a claude.ai Artifact that's a control-panel dashboard for my TRIGGERcmd connector — it lists all my TRIGGERcmd commands, grouped by computer, and lets me run any of them with an optional parameter, right from the page.

Before writing any code:
1. Load the artifact-capabilities skill and read the mcp.d.ts type definitions for the current runtime contract — follow that contract exactly, don't rely on remembered API shape.
2. Find my TRIGGERcmd MCP tools (they'll appear as mcp__<connector>__list_commands and mcp__<connector>__run_command or similar). Actually call list_commands, and actually call run_command once (e.g. trigger a harmless command like a Calculator command) to observe the REAL response shape — TRIGGERcmd tools return plain text, not JSON, so you'll need to parse that text yourself rather than assume structured JSON.
3. Note: the manifest `capabilities` you pass to the Artifact tool uses the connector name segment from the tool's internal name (e.g. "claude_ai_TRIGGERcmd"), which gets auto-resolved to the display name at publish time. BUT the actual `window.claude.mcp.callTool()` / `watchTool()` calls inside the page's JavaScript must use the connector's real DISPLAY NAME (e.g. "TRIGGERcmd"), not that internal segment — using the segment at runtime causes a `not_in_manifest` error even though the manifest itself is correct. Get this right the first time.

Dashboard requirements:
- Use watchTool for list_commands (it's a read, should stay live/refreshable) and callTool for run_command (it's an action, fire on click).
- Parse the text response into computer → [commands] groups; each command may have a voice-trigger label and/or a description (commands with descriptions typically accept a text parameter).
- Layout: sidebar or nav of computers with command counts, a search box that filters across all commands, collapsible groups per computer, and a Run button + optional parameter input per command.
- Show a live activity log (e.g. fixed panel) of triggered commands with pending/success/error status, since these are fire-and-forget hardware/device triggers with no persistent state to poll.
- Handle MCP failure states distinctly per the type definitions — don't collapse everything into one generic error banner: distinguish "connector not connected/needs reauth" (show a reconnect message) from "temporarily unavailable" (offer retry, keep last-good data visible) from "not available in this context" (full-page fallback, e.g. when window.claude.mcp is undefined).
- Design it as a real control-panel UI (dark/light theme aware, decent typography, not a default Bootstrap-y look) — apply the artifact-design skill's guidance for a UI/dashboard treatment.

Publish it as a shareable Artifact with an appropriate favicon.

Then you'll find this and other artifacts here: https://claude.ai/artifacts