Skip to content

Quick Start

The fastest way to build on Doover is to give your coding agent direct access to the platform. The Doover MCP is a hosted Model Context Protocol server that lets clients like Cursor, Claude Code, and Codex read your devices, work with live data, manage applications, and deploy to hardware --- all as you, with your normal permissions.

This page gets the MCP connected and shows you what to ask first. You need:

  • A Doover account. If your organisation has not set one up yet, contact support@doover.com.
  • An MCP client that supports remote HTTP servers and OAuth, such as Cursor, Claude Code, Codex, or VS Code.

You do not need to install the Doover command-line tool or the pydoover SDK to get started. Everything below is done through the MCP.

1. Connect the Doover MCP

Choose your client to manually set up Doover MCP. Every client connects to https://mcp.doover.com/ and uses your Doover account to sign in. There are no API keys to copy.

1. Add the server

Run this in your terminal. User scope makes Doover available in every project:

claude mcp add --transport http --scope user doover https://mcp.doover.com/

2. Sign in to Doover

Start Claude Code and run:

/mcp

Choose doover and follow the authentication prompt to sign in with your Doover account in the browser. Return to Claude Code and check that the server is connected.

See the Claude Code MCP documentation for more details.

Check your connection

Ask your agent:

Who am I signed in as on Doover?

Your agent should use the Doover MCP to return your name and email.

2. Ask your agent to read the guide

The Doover MCP ships with a built-in developer guide that explains how the platform fits together, which tool to reach for, and how apps are built and published. It is written for agents, and it is the source of truth --- when it disagrees with something the agent assumes, the guide is right.

Before your first real task, say:

Read the Doover developer guide before we start.

Your agent calls the guide tool, reads the table of contents, and opens the sections it needs as it goes.

3. Try it out

Some prompts to start with. Each one exercises a different part of the platform.

Look around your fleet

Show me the devices in my organisation and which are online.

Read live data from a device

What channels does the device called Pump Station 3 have? Show me the current values on its ui_state channel.

See what is installed on a device

Which apps are installed on Pump Station 3, and what versions are they running?

Find an existing app to reuse

Is there already a Doover app that reads a Modbus flow meter? Show me what is available to my organisation.

Checkmark Circle
Reads are safe

Listing devices, channels, apps and installs never changes anything. Your agent will ask before it does anything that writes to a device, such as sending a command or creating a deployment.

4. Build your first app

When you are ready to write software, your agent follows the workflow in the guide's app-dev section. In short:

  1. Start from a template. The getdoover/examples repository holds a minimal template for each app type, plus worked examples that add one behaviour at a time. Your agent copies the closest one.
  2. Develop locally. Templates use uv for dependencies, so uv sync and uv run pytest are how you install and test.
  3. Publish by pushing. Once your GitHub account or organisation is linked to Doover, a push to your repository's default branch releases the app through GitHub Actions. There is no publish command.
  4. Install and deploy through the MCP. Your agent creates an app install on a device with the right config, then creates a deployment to roll it out.

You need git, Python 3.11 or later, uv, and a GitHub account you can push to. Add Docker if you are building an app that runs on a device.

A good first prompt:

I have a Doovit on my bench with a temperature sensor on analog input 0. Read the Doover developer guide, then help me build a device app that reads the sensor once a second and shows it on the dashboard.

The agent will copy the device-app template, wire up the analog input, declare a tag and a UI variable, run the tests, and walk you through linking GitHub and deploying.

Optional: Doover skills for Claude Code

The doover-skills plugin adds detailed pydoover reference material and coding standards that Claude Code loads on demand while writing app code. It is a useful companion to the MCP once you are past the first app.

/plugin marketplace add getdoover/doover-skills
/plugin install doover-development@doover-skills

Troubleshooting

The doover server shows as disconnected or returns a 401. Try signing in again. In Claude Code run /mcp and choose doover. In Codex run codex mcp login doover. In Cursor or VS Code, use the server's reconnect or authentication action.

Calls fail with an error asking for an organisation. You belong to more than one Doover organisation and none is selected for this session. Ask your agent to list your organisations and select the one you want. The selection lasts for the session, so you may need to repeat it after reconnecting.

Claude Code does not list any Doover tools. Check the server is registered with claude mcp list. If it is missing, re-run the claude mcp add command from step 1. If you added it without --scope user, it is only available in the directory you were in at the time.

The agent created an app but cannot find it. It was almost certainly created in a different organisation. Ask the agent to list organisations, select the right one, and look again.

The agent starts installing the Doover CLI. It does not need to. Point it at the guide: the CLI is optional and only useful for running an app you are writing against a real device from a terminal, or for watching a channel live.

Next steps