About simplisafe-mcp
simplisafe-mcp is an MCP server for checking and controlling SimpliSafe 3 home security systems. It provides system, sensor, lock, settings and event information, and requires confirmation before sensitive actions such as changing alarm or lock state or reading PINs.
Key features
- 01View system and alarm status
- 02Check sensor and lock states
- 03Review events and system settings
- 04Arm or disarm the alarm
- 05Lock or unlock doors
- 06Confirm sensitive operations first
Use cases
- 01Check whether a system is armed or sensors are triggered
- 02Review recent alarm and door events
- 03Change alarm or lock state with a confirmation step
MCP support
Exposes SimpliSafe system, sensor, lock, event, settings and health-check tools to MCP clients. State-changing operations and PIN reads require confirmation before execution.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
npx -y simplisafe-mcpSIMPLISAFE_REFRESH_TOKENSecretOAuth refresh token from the one-time browser login (scripts/bootstrap-auth.mjs)SIMPLISAFE_TOKEN_CACHESecretSet to false to disable the on-disk token cache and re-authenticate on every process start. Defaults to enabled.SIMPLISAFE_TOKEN_FILESecretAbsolute path for the token cache file. Defaults to $MCP_DATA_DIR/.simplisafe-mcp/token.json.Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
10Saved tool list · updated 2w ago
This tool takes no parameters.
Details
From the project READMEsimplisafe-mcp
MCP server for SimpliSafe home security. Check whether the system is armed, review sensors and events, arm/disarm, and control smart locks — from Claude.
This server can disarm a home alarm and unlock doors. Every tool that changes physical state, plus the tool that reads alarm PINs, asks you to confirm first — nothing is sent until you do, and you see a preview of exactly what would happen (see Confirmations). Install it only where you'd be comfortable with that capability.
Developed and maintained by AI (Claude Code).
What you get
| Tool | |
|---|---|
simplisafe_list_systems | Systems on the account with current alarm state |
simplisafe_get_system | One system's state, connectivity, base-station messages |
simplisafe_list_sensors | Sensors with battery / offline / triggered status, filterable |
simplisafe_list_locks | Smart locks with locked / unlocked / jammed state |
simplisafe_get_events | Recent base-station events (arm, disarm, opens, alarms) |
simplisafe_get_settings | Entry/exit delays, volumes, base-station health |
simplisafe_get_pins | Alarm PINs — cleartext, asks you to confirm first |
simplisafe_set_alarm_state | Arm home / arm away / disarm — asks you to confirm first |
simplisafe_set_lock_state | Lock / unlock a door — asks you to confirm first |
simplisafe_healthcheck | Auth + API reachability |
Supports SimpliSafe 3 systems. Legacy SS2 systems are rejected with an explanation rather than an opaque upstream 404.
Install
npm install -g simplisafe-mcp
Or add to .mcp.json:
{
"mcpServers": {
"simplisafe": {
"command": "npx",
"args": ["-y", "simplisafe-mcp"],
"env": { "SIMPLISAFE_REFRESH_TOKEN": "${SIMPLISAFE_REFRESH_TOKEN}" }
}
}
}
Authentication — one browser login, once
SimpliSafe issues no API keys. The credential is an OAuth2 refresh token, minted by a browser login you perform one time:
git clone https://github.com/chrischall/simplisafe-mcp && cd simplisafe-mcp
node scripts/bootstrap-auth.mjs # prints an authorize URL
# sign in (MFA included), then copy the com.simplisafe.mobile:// URL
node scripts/bootstrap-auth.mjs "<that URL>"
The token is written to .env (mode 0600) after being verified against the live
API. SimpliSafe does not rotate refresh tokens, so it stays valid until you
sign out of all devices in the SimpliSafe app — which is how you revoke it.
Capturing the code: open DevTools → Network and tick Preserve log before
signing in; afterwards the browser fails to open a com.simplisafe.mobile://…
link, and that failed entry's link address is what you paste. The code is
single-use and expires in about two minutes.
Treat the resulting token like a house key: it grants full control of the alarm.
Confirmations
Arming, disarming, locking, unlocking and reading the PINs all ask you to
confirm first. On a client that can show a confirmation prompt (Claude Code) you
get the prompt, with the details below (unless MCP_CONFIRM_ELICITATION=off).
On one that cannot (claude.ai, Claude Desktop) the first call sends nothing and returns a preview plus a
confirmToken; only a repeat call with that token acts, once. The preview
includes a plain statement of the physical consequence:
{
"status": "confirmation-required",
"confirmed": false,
"dispatched": false,
"action": "alarm.set_state",
"preview": {
"action": "set alarm state to away",
"method": "POST",
"path": "/ss3/subscriptions/7858153/state/away",
"sid": 7858153,
"locationName": "Home",
"currentState": "OFF",
"requestedState": "AWAY",
"warning": "Arms ALL sensors including interior motion. Starts an exit delay; anyone still moving inside when it expires can trigger the siren and a monitoring-center dispatch."
},
"confirmToken": "…",
"expiresAt": "…",
"ttlSeconds": 600,
"instruction": "Show this preview to the user verbatim and proceed only after they explicitly approve in chat. Then call again with confirmToken."
}
The token is bound to that tool, target and preview. The system is re-read on
the second call, and if what would happen no longer matches what you approved —
different arguments, or the alarm or lock state moved in between — it is
refused (DRAFT_CHANGED) with a fresh preview. A token that was already used is
refused too (TOKEN_REUSED).
| variable | default | |
|---|---|---|
MCP_CONFIRM_MODE | ask-user | What a write does on a client that cannot show a confirmation prompt (claude.ai, Claude Desktop). ask-user: two steps — the first call does nothing and returns a preview plus a token, and the model must get your approval in chat before calling again with it. auto: the same two steps, but the model may use the token after reviewing the preview itself. refuse: writes are refused on such clients. A client that can show prompts (Claude Code) always gets the real prompt, unless MCP_CONFIRM_ELICITATION=off. An unrecognised value is treated as refuse. |
MCP_CONFIRM_ELICITATION | on | off never shows a confirmation prompt, so every client gets the MCP_CONFIRM_MODE behaviour. Set it for a client that says it can show prompts but never does (the write hangs — opencode 2.0.x). Any other value is treated as on, with a warning on stderr. |
MCP_CONFIRM_TTL_SECONDS | 600 | How long a token stays valid. |
MCP_CONFIRM_SECRET | random per process | Signing key; set it only if tokens must survive a server restart. |
Writes are verified
Once confirmed, a write executes and then re-reads the system to check what
actually happened, reporting confirmed, in_progress (the exit delay is
counting down), or unconfirmed. A 2xx is never treated as proof.
Shell access without the server
For quick one-off queries there's a curl-based skill in
skills/simplisafe-api/ — same API, no MCP
process, sharing the same refresh token.
Development
npm install
npm run build
npm test
Verified endpoint shapes live in docs/SIMPLISAFE-API.md, including several things that are easy to get backwards:
- lock state is encoded 1 = locked, 2 = unlocked;
- system version for routing is at
location.system.version, not the top-levelsystemVersion; - events and doorlock control are not under the
ss3/prefix; numEventshas an undocumented hard ceiling of 50;settings.pinsreturns alarm codes in cleartext alongside harmless settings.
Disclaimer
Unofficial. Not affiliated with or endorsed by SimpliSafe. It uses the same private API the SimpliSafe mobile app uses, with your own account credentials. Use at your own discretion.
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.