What do I need to prepare before deploying?

Three things. A SenseCAP platform Access ID and Access Key, an MCP access-point URL from the XiaoZhi console, and a host with Docker installed. The first deploy also needs GitHub access to build the image.

Does it work offline?

No. The bridge fetches live data from the SenseCAP cloud platform and connects to the assistant's access point, so both sides need the internet.

Where does the data go, and do I need to open a port?

Device data flows from the SenseCAP platform through the bridge host to the assistant access point you configure. The bridge only dials out and listens on no port, so no firewall change is needed. Credentials are entered at deployment and never written into the image.

Can it connect to assistants other than XiaoZhi?

Yes. Any MCP-capable host works, such as Claude Desktop, Cursor or your own LLM agent.

The spoken times are wrong. How do I fix that?

Enter your devices' time zone, such as Asia/Shanghai, in the Device Timezone deploy field and redeploy. Left blank, the China site uses Beijing time and other sites use UTC; it never guesses.

Can I add my own tools or change what gets spoken?

Yes. The bridge source is public on GitHub, where the tool definitions and spoken text live; build your own image from your changes and deploy it.

What should we check on licensing before commercial use?

The bridge source is used under its GitHub repository licence, the SenseCAP platform API under your account terms, and the voice assistant or AI host under its own terms of service. Check each before shipping a closed product.

SenseCAP PaaS MCP Bridge

Give any MCP-compatible voice assistant or LLM a spoken, anomaly-first report on your SenseCAP PaaS device fleet — ask "how are my devices doing" or "what's the greenhouse reading right now" and get a real answer, no app or dashboard needed.

SenseCraftBeginner10minInteractive Smart DisplayAI Agent ReadySource Code

reComputer R1125-10 from the reComputer R1100 Series, the reference host for the bridge container

Reusable Site Templates

Sites where devices already report to the SenseCAP platform, and people want to ask a question out loud instead of opening an app or dashboard.

Farm and greenhouse rounds — ask "how is the greenhouse weather station doing" and hear current temperature, humidity, soil moisture, wind and rainfall.
Daily fleet checks — ask "how are my devices doing" and hear who is offline or low on battery first; a healthy fleet gets one sentence.
Device data for an AI assistant — Claude Desktop, Cursor or your own agent looks up live and historical data by device name.

Not a fit: devices not yet connected to the SenseCAP platform. It reads data already on the platform and does not talk to sensors directly.

How it works

Three pieces: SenseCAP devices already reporting to a platform account, a host running the bridge container, and a voice assistant or AI that uses MCP tools. The bridge reads from the platform and dials out to the assistant's access point, with no inbound port to open.

Devices keep reporting to the SenseCAP platform as usual; the bridge only reads what is on the platform.

Needs a platform API key: create an Access ID and Access Key on the API keys page at sensecap.seeed.cc.
Pick the site: choose the global or China site at deployment; the time zone can be set separately.

Any always-on machine that runs Docker works; the reference host is a reComputer R1100 Series.

Item	Detail
Host	reComputer R1100 Series, or any Docker host
Network	Reaches the SenseCAP platform and the assistant access point; the first deploy builds the image from GitHub source
Ports	Listens on none; connects outward only
Credentials	Entered at deployment, never baked into the image; redeploy to rotate them

The main target is the XiaoZhi voice assistant; any MCP-capable host also works.

Host	How to connect
XiaoZhi voice assistant	Copy the MCP access-point URL from the XiaoZhi console and enter it at deployment
Claude Desktop, Cursor, your own agent	Follow the host's MCP configuration

The data interface is listed under What You Get below.

How well it works

The table lists the 9 tools the assistant can call after deployment, in three groups.

Group	What it does	Limit
Spoken (fleet overview, one-device reading)	Reads out a ready-made sentence; the overview leads with anomalies	Reports only the latest data already on the platform
Data (live, history, chart aggregation, device key)	Returns a short summary plus the full structured data	Built for scripts and follow-up processing, not for reading aloud
Setup (register device, code templates)	Registers a device by name and returns its EUI; browses Arduino/PlatformIO reporting templates	Templates target Arduino and PlatformIO projects

Times are spoken in the time zone set at deployment; left blank, the China site uses Beijing time and other sites use UTC.

What you get out of it

Once deployed, this is the connection.

Exit	What comes out	Notes
Outbound WebSocket to the MCP access point	9 MCP tools for the assistant to call	The bridge dials out; no port opens on the host

If the container keeps restarting, check its logs; the usual cause is a wrong Access ID/Key or access-point URL. If the assistant never calls a tool, first check the access-point URL was copied in full, including the token at the end.

How to deploy

Check the setup and its product list, then follow the deployment guide.

Hardware Sizing & Deployment Setup

Select site conditions to generate recommended bundles and guides, or let an AI assistant estimate hardware and deploy for you.

Self-service Consult & Deploy

Deployment Options