shelly-mcp

shelly-mcp

MCP server for the entire Shelly smart-home ecosystem, enabling local-first control, automation, and energy monitoring of Shelly devices with cloud fallback.

Category
Visit Server

README

shelly-mcp

mcp-name: io.github.Buggy1111/shelly-mcp

MCP server for the entire Shelly smart-home ecosystem — read, control, and automate Shelly devices of every generation (Gen1 → Gen4; BLU via its gateway or the generic RPC engine, dedicated BLU tools are on the roadmap) from any MCP client (Claude Desktop, Claude Code, Cursor, …). Local-first (zero rate-limit, ~10 ms, full API), with cloud fallback for off-LAN access.

⚠️ Unofficial community project. Not affiliated with, endorsed by, or sponsored by Allterco Robotics / Shelly. "Shelly" is a trademark of its respective owner.

🚧 Alpha / in development. See docs/ for the full design.

Why

The only existing Shelly MCP servers are cloud-only and minimal. shelly-mcp unifies Gen1 and Gen2+ behind one tool surface, covers energy monitoring and automation (schedules, scripts, webhooks, KVS), and auto-discovers any device's capabilities — including hardware released after this server was written.

Install

uvx shelly-mcp          # or: pip install shelly-mcp

Register in your MCP client:

{ "mcpServers": { "shelly": { "command": "uvx", "args": ["shelly-mcp"] } } }

Configure

Auto-discovery (mDNS) finds devices on your local subnet — shelly_discover and you're running. For named devices ("turn off the kitchen"), devices on other subnets, passwords, or the cloud fallback, create ~/.config/shelly-mcp/config.yaml (chmod 600):

devices:
  televize:
    ip: 192.168.0.101
    location: obývák            # lets "turn off the living room" work
    # password: "..."           # only if the device has auth (or env SHELLY_PW_televize)
cloud:
  enabled: false                # optional off-LAN fallback
  # auth_key via env SHELLY_CLOUD_AUTH_KEY

Full reference: config.example.yaml and docs/04-CONFIG-AND-DEPLOY.md.

What data leaves your machine

None, beyond the calls to your own Shelly devices (on your LAN) and — only if you enable it — your own Shelly Cloud account. No telemetry, no phone-home.

Capabilities (local vs cloud)

Local Cloud
Discovery, status, control ✅ ⚠️ control + status only
Energy live + history ✅ ⚠️ live only
Automation (schedules/scripts/webhooks/KVS) ✅ ❌

Tools, resources & prompts

Read (safe): shelly_version · shelly_discover · shelly_list_devices · shelly_get_info · shelly_get_status (normalized) · shelly_get_config (credentials masked) · shelly_list_components · shelly_list_methods

Control (audited): shelly_switch_set · shelly_switch_toggle · shelly_light_set (RGBW/CCT/white) · shelly_cover_move

Energy: shelly_energy_live · shelly_energy_history

Generic engine (total coverage): shelly_rpc (read-only) · shelly_rpc_write (mutations, confirm:true + data-loss double-gate)

System / schedules (gated): shelly_system_reboot|update|set_auth · shelly_schedule_list|create|update|delete

Automation (Gen2+ local-only): shelly_kvs_* (key-value store) · shelly_webhook_* (event→HTTP) · shelly_script_* (on-device JS — list/get_code/create/put_code/start/stop/eval/delete, chunked upload) · shelly_virtual_* (virtual components). Deletes + arbitrary-code paths (script_put_code/eval) are confirm:true-gated.

Scenes (deterministic, named): shelly_scene_list|get|run|create|delete — define a multi-device routine once and run it by name (shelly_scene_run "film"), identical every time and schedulable from any client. Stored in ~/.config/shelly-mcp/scenes.yaml (see scenes.example.yaml); scenes and schedules accept only plain control methods (Switch/Light/RGB(W)/CCT/Cover) — never Script.Eval, SetAuth, or anything destructive (ADR-007, docs/06-SCENES.md).

Resources: shelly://devices, shelly://device/{name}/status — Prompts: shelly_evening_scene, shelly_energy_report, shelly_diagnose

Safety: reads are readOnlyHint; every mutation is audit-logged; the generic write tool and destructive system tools require explicit confirm:true, and irreversible methods (factory reset, wipe-all) need a second i_understand_data_loss gate — so even a hijacked LLM can't silently destroy a device.

Troubleshooting

  • "Device unreachable" — confirm the IP (shelly_discover, your router's client list, or the Shelly app → device → Settings → Device information), and that the machine running the server is on the same LAN. In WSL/containers, mDNS discovery usually doesn't work — configure devices by ip in the config file instead (that path needs no mDNS).
  • "Auth required" — the device has a password: add password: under the device in the config (or SHELLY_PW_<name> env var).
  • Cloud-only device refuses automation tools — expected: the Shelly Cloud API can't manage schedules/scripts/webhooks/KVS; connect locally for those.
  • Config refuses to load — if it contains a secret, it must be chmod 600 (deliberate, fail-closed).

Docs

Full design in docs/ (indexed): overview, architecture (+ADRs), tool surface, security, scenes, config/deploy, build plan, project log, roadmap, the launch runbook, and the complete Shelly API catalog.

License

MIT — see LICENSE.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured