tapo-mcp
MCP server that exposes Tapo smart-home devices as AI-callable tools and resources via the Model Context Protocol.
Built on the tapo crate and the rmcp SDK. Runs as an HTTP server (Streamable HTTP transport).
Example Prompts
"List all my Tapo devices"
"Turn off the office light"
"Turn on smart plug 4 on the power strip"
"Is the office light on?"
"Set the bedroom light to 50% brightness"
"Change the living room light to Coral"
"Take a snapshot from the baby monitor"
"What's the temperature in the kitchen?"
"Show me the last 24 hours of temperature from the living room temperature and humidity sensor"
"List the 5 most recent events on the smart button"
"How much energy did the washing machine plug use each day this month?"
"Show me the TV's power draw over the last 6 hours"
| Tool | Description |
|---|
list_devices | List available Tapo devices on the network (includes set and get capabilities). |
get_device_state | Get a device's current state (e.g. {"type": "DeviceInfo"}). Checks the capability's parameters, then verifies the device ID matches at the given IP. |
control_device | Control a device by applying 1 to 12 set capabilities, in order. Checks the count and values, then verifies the device ID matches at the given IP; a capability the device does not support fails before any is applied. Returns the device's state afterwards, read back from the device. |
take_snapshot | Capture a still JPEG snapshot from a Tapo camera (~640x360), with the device id, capture time and size as text, plus a short-lived link when TAPO_MCP_PUBLIC_URL is set (see Snapshot links). Verifies the device ID matches at the given IP first. |
Resources
| URI | Description |
|---|
tapo://devices | JSON list of discovered Tapo devices. |
Capabilities
Devices and child devices expose separate lists of set and get capabilities they support. Child devices have no IP of their own; their ip is their parent's, which is the one to pass with the child's id.
Set Capabilities
| Capability | Description |
|---|
Brightness | Set the device brightness (1-100) |
Color | Set the device color using a preset name |
HueSaturation | Set the device color using a hue (0-360) and saturation (1-100) |
ColorTemperature | Set the device color temperature (2500-6500) |
OnOff | Turn the device on or off |
Get Capabilities
| Capability | Description |
|---|
Detections | Read the detections stored on a camera's SD card, with the events detected (person, motion, ...), for up to 24 hours (C120, C210, C220, C225, C325WB, C520WS, TC40, TC70) |
DeviceInfo | Read the device's current state |
EnergyDataHourly | Read the energy used per hour, in Wh, for up to 8 days (P110, P110M, P115, P304M, P316M) |
EnergyDataDaily | Read the energy used per day, in Wh, for a quarter of a year (P110, P110M, P115, P304M, P316M) |
EnergyDataMonthly | Read the energy used per month, in Wh, for a year (P110, P110M, P115, P304M, P316M) |
EnergyMeterData | Read the live voltage, current and power, and the energy used since the device last restarted (P110, P110M, P115, P304M, P316M) |
PowerDataEvery5Minutes | Read the power draw every 5 minutes, in W, for up to 12 hours (P110, P110M, P115, P304M, P316M) |
PowerDataHourly | Read the power draw every hour, in W, for up to 6 days (P110, P110M, P115, P304M, P316M) |
Snapshot | Capture a still JPEG snapshot. Served by the dedicated take_snapshot tool (binary, not JSON state) |
TemperatureHumidityRecords | Read the last 24 hours of temperature and humidity records (T310, T315) at 15 minute intervals |
TriggerLogs | Read paginated trigger logs from a hub child sensor (S200, T100, T110, T300) |
Configuration
All configuration is via environment variables prefixed with TAPO_MCP_:
| Variable | Required | Default | Description |
|---|
TAPO_MCP_USERNAME | Yes | β | Tapo account email |
TAPO_MCP_PASSWORD | Yes | β | Tapo account password |
TAPO_MCP_CAMERA_USERNAME | No | β | Camera account username1. Required by take_snapshot. |
TAPO_MCP_CAMERA_PASSWORD | No | β | Camera account password1. Required by take_snapshot. |
TAPO_MCP_DISCOVERY_TARGET | Yes | β | IP address for device discovery, usually the network's broadcast address (e.g. 192.168.1.255) |
TAPO_MCP_HTTP_ADDR | No | 127.0.0.1:3000 | Address the server listens on |
TAPO_MCP_DISCOVERY_TIMEOUT | No | 5 | Discovery timeout in seconds. Must be between 1 and 602 |
TAPO_MCP_DISCOVERY_INTERVAL | No | 600 | Seconds between background discoveries that refresh the device list. Must be at least TAPO_MCP_DISCOVERY_TIMEOUT and at most 86400 (one day)2 |
TAPO_MCP_API_KEY | No | β | Bearer token for HTTP authentication (see below) |
TAPO_MCP_ALLOWED_HOSTS | No | loopback only | Comma-separated Host header allowlist (see Network exposure) |
TAPO_MCP_PUBLIC_URL | No | β | Base URL clients reach the server at (e.g. https://tapo.example.com). Enables short-lived snapshot links (see Snapshot links) |
Clients connect to the MCP endpoint at /mcp on that address, for example http://127.0.0.1:3000/mcp. / still works but is deprecated: it logs a warning and will be removed in a later release.
Authentication
When TAPO_MCP_API_KEY is set, the server requires every HTTP request except /livez and snapshot links to include an Authorization: Bearer <key> header. Requests with a missing or invalid token receive a 401 Unauthorized response.
When the variable is unset (or empty/whitespace-only), the server runs without authentication.
Network exposure
The server enforces the MCP Streamable HTTP DNS-rebinding protection. By default only loopback Host headers (localhost, 127.0.0.1, ::1) are accepted, which prevents a malicious web page from reaching a locally running server via DNS rebinding. Requests to the MCP endpoint with any other Host receive a 403 Forbidden response, and the server logs a warning that names the refused host. Snapshot links are not covered by this check (see Snapshot links).
To reach the server over the LAN or from another host, set TAPO_MCP_ALLOWED_HOSTS to the exact hostname(s) or host:port authorities clients connect to, for example TAPO_MCP_ALLOWED_HOSTS="tapo-mcp.lan:3000,192.168.1.50:3000". This replaces the loopback default, so include loopback entries as well if you still need them.
To avoid shipping unauthenticated smart-home control, the server refuses to start when it binds to a non-loopback address (for example 0.0.0.0:3000) without TAPO_MCP_API_KEY set. Set an API key, or bind to a loopback address.
Snapshot links
Some MCP clients don't pass image results on to the model, and others show the image to the model but give it nothing it can hand to the user. When TAPO_MCP_PUBLIC_URL is set, take_snapshot also returns a url (and its expires_at) pointing at <TAPO_MCP_PUBLIC_URL>/snapshots/<token>.jpg, so the agent can share the link or download the file.
- The link is served without the API key, so a browser can open it directly. The
Host allowlist from TAPO_MCP_ALLOWED_HOSTS does not apply to it either. The random token in the path is the only credential: anyone holding the link can view that frame until it expires.
- Each link is valid for 5 minutes. Snapshots are kept in memory only and never written to disk. Expired snapshots are cleared out within a minute, and at most the 1024 most recent are kept.
- Behind an authenticating reverse proxy (Cloudflare Access and similar), links opened in a browser that is already signed in to the proxy work. Non-browser fetches (
curl, an agent downloading the file) need the proxy's own credential or a policy exception for /snapshots/.
When TAPO_MCP_PUBLIC_URL is unset, no link is returned and the /snapshots/ route is not served.
OpenTelemetry
To export traces, set OTEL_EXPORTER_OTLP_ENDPOINT to the base URL of an OTLP/HTTP collector (e.g. http://alloy.monitoring:4318); the server sends them to /v1/traces under it. To send them to another path, set OTEL_EXPORTER_OTLP_TRACES_ENDPOINT to the full URL instead. OTEL_RESOURCE_ATTRIBUTES adds resource attributes. RUST_LOG (default tapo_mcp=info) also decides which spans are exported.
What gets traced:
- Tool calls. Each
tools/call request gets one server span named tools/call <tool>, covering the whole call, following the OpenTelemetry MCP semantic conventions. It carries mcp.method.name, gen_ai.tool.name (only for tools the server provides), gen_ai.operation.name, jsonrpc.request.id, mcp.session.id and mcp.protocol.version, plus tapo.device.id and tapo.device.ip where the tool takes them. A failed call has error status and error.type (the JSON-RPC error code, or tool_error). Inside it, the check_device span covers the verification that the device ID matches at the given IP, and get_capability, set_capability or capture_snapshot covers the device request. get_capability and set_capability carry tapo.capability, and control_device has one set_capability span for each capability it applies, then a get_capability span for the device info it reads back. A failed step has error status and error.type set to the kind of failure, such as DeviceNotFound or WrongDeviceType.
- Background jobs.
refresh_devices for each device discovery, and prune_snapshots for each sweep of expired snapshot links.
Credentials, the API key and image data are never recorded. Tool arguments are not recorded either, except id and ip, and a failed call's error message, which can repeat the value that was rejected.
Incoming W3C trace context is honoured. When the request's params._meta has a traceparent (and optionally tracestate), as MCP clients that trace their calls send, the tool call span becomes its child and joins that trace. A traceparent header on the HTTP request that carries the call, for example from a tracing reverse proxy, is then recorded as a span link. Without _meta context, the HTTP header is used as the parent instead. Without either, the span starts a new trace.
Deployment
Docker
docker run --rm \
--network host \
-e TAPO_MCP_USERNAME="you@example.com" \
-e TAPO_MCP_PASSWORD="<YOUR_TAPO_PASSWORD>" \
-e TAPO_MCP_CAMERA_USERNAME="<YOUR_CAMERA_ACCOUNT_USERNAME>" \
-e TAPO_MCP_CAMERA_PASSWORD="<YOUR_CAMERA_ACCOUNT_PASSWORD>" \
-e TAPO_MCP_DISCOVERY_TARGET="192.168.1.255" \
-e TAPO_MCP_API_KEY="<YOUR_TAPO_MCP_API_KEY>" \
ghcr.io/mihai-dinculescu/tapo-mcp:latest
Note: The image binds to 0.0.0.0:3000, so TAPO_MCP_API_KEY is required β the server refuses to start on a non-loopback address without it (see Network exposure). To reach the server by hostname or LAN IP rather than loopback, also set TAPO_MCP_ALLOWED_HOSTS. To get snapshot links, add -e TAPO_MCP_PUBLIC_URL="https://tapo.example.com".
Note: --network host is required so the container can reach Tapo devices on your local network via UDP broadcast for discovery. On macOS and Windows, --network host is not supported β you can use -p 3000:3000 instead, but device discovery won't work as Docker Desktop runs containers inside a VM without LAN access.
Kubernetes
Create the Secret and ConfigMap first:
kubectl create secret generic tapo-mcp-secrets \
--from-literal=TAPO_MCP_USERNAME="you@example.com" \
--from-literal=TAPO_MCP_PASSWORD="<YOUR_TAPO_PASSWORD>" \
--from-literal=TAPO_MCP_CAMERA_USERNAME="<YOUR_CAMERA_ACCOUNT_USERNAME>" \
--from-literal=TAPO_MCP_CAMERA_PASSWORD="<YOUR_CAMERA_ACCOUNT_PASSWORD>" \
--from-literal=TAPO_MCP_API_KEY="<YOUR_TAPO_MCP_API_KEY>"
kubectl create configmap tapo-mcp-config \
--from-literal=TAPO_MCP_DISCOVERY_TARGET="192.168.1.255"
Then apply the Deployment:
apiVersion: apps/v1
kind: Deployment
metadata:
name: tapo-mcp
spec:
replicas: 1
strategy:
type: Recreate
selector:
matchLabels:
app: tapo-mcp
template:
metadata:
labels:
app: tapo-mcp
spec:
hostNetwork: true
containers:
- name: tapo-mcp
image: ghcr.io/mihai-dinculescu/tapo-mcp:latest
env:
- name: TAPO_MCP_USERNAME
valueFrom:
secretKeyRef:
name: tapo-mcp-secrets
key: TAPO_MCP_USERNAME
- name: TAPO_MCP_PASSWORD
valueFrom:
secretKeyRef:
name: tapo-mcp-secrets
key: TAPO_MCP_PASSWORD
- name: TAPO_MCP_CAMERA_USERNAME
valueFrom:
secretKeyRef:
name: tapo-mcp-secrets
key: TAPO_MCP_CAMERA_USERNAME
- name: TAPO_MCP_CAMERA_PASSWORD
valueFrom:
secretKeyRef:
name: tapo-mcp-secrets
key: TAPO_MCP_CAMERA_PASSWORD
- name: TAPO_MCP_API_KEY
valueFrom:
secretKeyRef:
name: tapo-mcp-secrets
key: TAPO_MCP_API_KEY
- name: TAPO_MCP_DISCOVERY_TARGET
valueFrom:
configMapKeyRef:
name: tapo-mcp-config
key: TAPO_MCP_DISCOVERY_TARGET
livenessProbe:
httpGet:
path: /livez
port: 3000
periodSeconds: 30
resources:
requests:
cpu: 10m
memory: 32Mi
limits:
memory: 256Mi
Note: hostNetwork: true is required for UDP broadcast discovery, similar to --network host in Docker. Because clients reach the server by node IP or hostname rather than loopback, set TAPO_MCP_ALLOWED_HOSTS accordingly (see Network exposure). To get snapshot links, add TAPO_MCP_PUBLIC_URL to the ConfigMap and the Deployment's env.
Health checks
GET /livez answers 200 OK while the server is running. It needs neither the API key nor a Host from TAPO_MCP_ALLOWED_HOSTS, so a liveness probe works as it is.
The server uses about 15 MiB of memory. Snapshots for snapshot links are kept in memory for 5 minutes, which is what the 256 MiB limit above leaves room for.
OpenClaw
The tapo skill for OpenClaw makes it easy to use a deployed tapo MCP server from OpenClaw agents.
Install it with:
Contributing
Contributions are welcome and encouraged! See /tapo-mcp/CONTRIBUTING.md.