wled-mcp

MCP server for controlling WLED LED controllers. Discovers devices on the local network via mDNS and exposes tools for managing state, presets, effects, and schedules.

Watch the demo — click the image below to see it in action.

Running

Public (no auth)

cargo run

OAuth

AUTH_TYPE=oauth OAUTH_CLIENT_ID=my-client OAUTH_CLIENT_SECRET=my-secret cargo run

The OAuth flow is fully stateless — tokens and auth codes are HMAC-signed with the client secret, so there is no server-side session state. You can restart or scale out without invalidating tokens.

Authorization auto-approves (no user interaction). Access and refresh tokens expire after 365 days.

Environment Variables

Variable	Description	Default
AUTH_TYPE	public or oauth	public
OAUTH_CLIENT_ID	Required when AUTH_TYPE=oauth	--
OAUTH_CLIENT_SECRET	Required when AUTH_TYPE=oauth	--
OAUTH_ALLOWED_REDIRECT_URIS	Comma-separated allowlist of redirect URIs. Patterns ending in * are prefix matches.	See below
BIND_ADDRESS	Address to bind	0.0.0.0:8080
RUST_LOG	Log level	wled_mcp=info

Default OAUTH_ALLOWED_REDIRECT_URIS:

https://chatgpt.com/connector/oauth/*,https://claude.ai/api/mcp/auth_callback,https://claude.com/api/mcp/auth_callback

The server derives its public URL from request headers (X-Forwarded-Proto, X-Forwarded-Host, Host), so no explicit base URL configuration is needed when running behind a reverse proxy like Caddy or nginx.

Endpoints

Path	Description
/	Landing page
/mcp	MCP endpoint (point Claude here)
/.well-known/oauth-authorization-server	OAuth metadata (oauth mode)
/.well-known/oauth-protected-resource	Protected resource metadata (oauth mode)
/oauth/authorize	Authorization endpoint (oauth mode)
/oauth/token	Token endpoint (oauth mode)

Future Improvements

Contributions welcome! Here are some areas that could use attention:

• Docker image / cargo install support — make it easier to run without cloning the repo and building from source.
• Rate limiting — the OAuth endpoints have no built-in rate limiting. Fine behind a reverse proxy that handles it, but a standalone deployment could benefit from this.
• Effects — experiment with having MCP set effects on patterns (it might be ready for prime time, but I haven't played with it).

License

MIT

---

This project was made possible by Mozr, where we help businesses design and implement AI-powered integrations and workflows. Have a project in mind? Let's talk.