AI Integration · Native MCP Server

Talk to your infrastructure. It listens — and acts.

Widgelix connects your devices to AI assistants like ChatGPT and Claude. Ask in plain language to build a dashboard, set an alert, or check what's happening in a building — and the AI does it. No console to master, no documentation to read.

Works with any MCP-compatible AI assistantPermission-controlledFully audited
/ The idea

Most IoT platforms make you learn them.
This one you just talk to.

Every IoT platform has a learning curve — menus, schemas, automation builders, decoder formats. Widgelix removes it. Because it exposes the whole platform through a native MCP server — the open standard that lets AI assistants connect to real systems — a connected AI can both run your infrastructure and set it up for you, from a plain-language request. You bring the intent. The AI handles the mechanics.

/ Capabilities

Two things your AI can do with Widgelix

Configure, by conversation

Configure, by conversation

Set up the platform without touching the console. Describe what you want, and the AI builds it: device types and decoders, dashboards, and automation actions — all from natural language.

Example

“Connect this air-quality sensor, show CO₂ and temperature on a dashboard, and alert me if CO₂ stays over 1000 ppm for 15 minutes.”

The AI creates the decoder, builds the dashboard, and writes the action. Ready in minutes.

Operate, in real time

Operate, in real time

Once it's running, the AI becomes an operator. It reads live and historical telemetry across every device, interprets what's happening, and takes action — within the permissions you set.

Example

“Which sites had abnormal energy use this week, and shut down the overnight load on the worst one.”

The AI analyzes the data, names the site, and sends the downlink command — with a logged audit trail.

/ How agents work

Read. Reason. Act.

Read

Agents access real-time and historical data across all your connected devices.

Reason

Agents interpret conditions, spot anomalies, and decide what needs to happen.

Act

Agents trigger actions and send downlink commands to devices — strictly within the limits you define.

/ What agents can do

The MCP server, area by area

Agents get structured tools, not screen scraping. These are the main tool areas of the Widgelix MCP server, plus ready prompts and documentation resources written for agents.

Assets & templates

List, read, create and update assets and asset templates, read telemetry and the last payload, and test uplink and downlink codecs before saving.

Actions

Create, update, clone, enable and disable actions, and check them for errors.

Dashboards

List, read, create and update site dashboards and their widgets.

Events & incidents

Read organisation and asset events, analyse incidents, and acknowledge or resolve them.

Agent memory

Read, add, update, archive and delete stored context about the organisation, templates and assets.

Docs & schemas

Read the platform documentation and the schemas for actions, dashboards, templates and assets, so the agent builds valid configuration.

/ Agent memory

Agents that remember your sites

Widgelix keeps a persistent AI context: facts about your organisation, each asset template and each asset. Location, purpose, installation details, calibration baselines, expected behaviour.

Agents read the context before they work with a device, ask a few short questions when something is missing, and save the facts you confirm. Outdated facts are archived instead of overwritten, so the next session, with any agent, starts informed.

  • Three levels: organisation, asset template, asset
  • Categories: environment, purpose, installation, calibration, behaviour, other
  • Pinned entries, tags and archiving
  • Each entry records whether an agent, the system or an import wrote it
/ Connect

Connected in minutes

01

Create an MCP user

In Settings → MCP Server, an organisation admin adds an MCP user, for example “Claude integration”, and copies the access token.

02

Add the server to your client

Use Streamable HTTP (or SSE for older clients) and authenticate with the token, or sign in through OAuth 2.1 with PKCE.

03

Ask

“Which freezers were warm last night?” or “Create a dashboard for building B.” The agent uses the tools above to answer and build.

/ Architecture

Built on MCP, the open standard for AI-to-system connections

Widgelix ships with a native Model Context Protocol (MCP) server. MCP is the open standard that lets AI assistants securely connect to external systems and act on them. Any MCP-compatible client — ChatGPT, Claude, or your own agent — connects directly to your Widgelix environment and gains structured access to your devices, dashboards, and actions.

There's no middleware to build and no custom integration per assistant. Connect the MCP server once, and your infrastructure becomes something an AI can read and operate through a single, standard interface.

Open standard

MCP-compatible by design; not locked to one AI vendor.

Structured access

Agents see devices, telemetry, dashboards, and actions as first-class objects, not scraped text.

Two-way

Agents don't just read; they can send downlink commands and build configuration.

/ Trust

You stay in control. Always.

Letting an AI act on physical infrastructure only works if the guardrails are real. They are.

Role-based permissions

Every agent operates inside the exact scope you grant. No access you didn't give.

Full audit trail

Every read and every action an agent takes is logged, attributable, and reviewable.

Bounded actions

Agents can only trigger what your actions and permissions allow — not arbitrary commands.

Human-in-the-loop, when you want it

Keep sensitive actions gated behind approval.

/ In practice

What teams actually do with it

Facility managers

Ask plain-language questions across a whole portfolio ("which buildings are coldest right now?") instead of clicking through dashboards.

System integrators

Stand up a new client's devices, dashboards, and alerts by describing them, cutting setup time per project.

Operations teams

Let an agent watch telemetry around the clock and act on anomalies the moment they cross a threshold.

Developers

Connect a custom agent to the MCP server and build automation on top of structured, permissioned access.

See your infrastructure answer back.

Bring a device and a question. We'll show you how Widgelix lets an AI assistant set it up, read it, and act on it — live, in a 30-minute walkthrough.

  • Live demo with ChatGPT or Claude connected to real devices
  • A look at the permission and audit model
  • Setup-by-conversation, end to end
/ FAQ

Frequently asked

+What is MCP?

The Model Context Protocol — an open standard that lets AI assistants connect to and act on external systems. Widgelix includes a native MCP server, so any MCP-compatible AI can work with your infrastructure.

+Which AI assistants work with it?

Any MCP-compatible client, including ChatGPT and Claude, as well as custom agents you build yourself.

+Can the AI actually control devices, or only read data?

Both — within your permissions. Agents read telemetry and can send downlink commands to actuate devices, always inside the scope and audit controls you define.

+Is it safe to let an AI act on my infrastructure?

Every agent action is permission-scoped and logged. Agents can only do what your roles and actions allow, and you can keep sensitive operations behind human approval.

+Do I need to write code?

No. Configuration and operation happen through natural-language requests to a connected assistant. Developers can build directly against the MCP server if they want deeper automation.

+How do I connect an AI assistant?

An organisation admin opens Settings → MCP Server and adds an MCP user. Connect your client to the Widgelix MCP server with its access token, or sign in through OAuth 2.1 (used by Claude and other clients that support it).

+What is agent memory?

Widgelix stores facts about your organisation, asset templates and individual assets, such as location, purpose, installation details or calibration. Agents read this context before working with a device and save new facts you confirm, so the next session starts informed.