ha-direct-access — Claude Skill for Home Assistant (claude.ai Projects + Desktop Commander)
Hey everyone! Wanted to share something I’ve been building that fills a gap I couldn’t find a solution for.
The problem
Every existing Claude + HA integration I found requires either Claude Code (a separate CLI subscription) or a custom MCP server running on your network. But I use claude.ai Projects day-to-day — not Claude Code — and I didn’t want to run extra infrastructure just to let Claude touch my HA config.
What this is
ha-direct-access is a Claude skill for claude.ai Projects users with Desktop Commander connected. It gives Claude full autonomous access to your Home Assistant instance — directly from your claude.ai chat, with no extra servers, no Claude Code required.
Once installed, Claude can:
- Edit
automations.yaml,scripts.yaml,configuration.yamldirectly over SSH - Call any HA REST API service
- Reload automations, scripts, HomeKit Bridge — no full restarts
- Run config validity checks and scan for repairs before closing any task
- Debug automations using traces and HA logs
- Update and repackage the skill itself as your setup evolves
How it’s different from existing solutions
| This Skill | Claude Code Skills | HA MCP Server (official) | ha-mcp | |
|---|---|---|---|---|
| Interface | claude.ai web/app | Claude Code CLI | Claude Desktop | Claude Desktop / API |
| Requires Claude Code |  No |
 Yes |
No |
No |
| Requires server on HA | No |
No |
Yes |
Yes |
| Edits config files directly | Yes |
Yes |
No |
No |
| Works in claude.ai Projects | Yes |
No |
No |
No |
| Self-updating skill file | Yes |
No |
No |
No |
Prerequisites
- claude.ai Pro (Projects required)
- Desktop Commander MCP connected (desktopcommander.app)
- Python 3 +
paramiko(pip3 install paramiko) - HA SSH & Terminal add-on running on port 22
- HA Long-lived access token
- Context7 MCP optional but recommended (prevents deprecated YAML syntax)
Quick start
git clone https://github.com/Bank1824/ha-direct-access-skill.git
cd ha-direct-access-skill
./ha-direct-access/configure.sh
Then package and upload to claude.ai → Settings → Skills. Full instructions in SETUP.md.
The skill grows with your setup
The SKILL.md lives on your local machine and acts as a persistent reference — entity IDs, Z2M friendly names, MQTT patterns, common gotchas. At the end of any session you can ask Claude to update it with anything new discovered. It stays current without manual maintenance.
Happy to answer questions. This came out of building a fairly deep local-first smart home setup and wanting Claude to be able to do the heavy lifting autonomously. Would love feedback from others who’ve tried similar setups.