Skip to Content
Get StartedAgent PluginOverview

Install the Arcade agent plugin

The Arcade agent plugin puts Arcade inside your coding agent with one command. Your agent gains tools for email, calendar, chat, issues, docs, and dozens of other apps, and you authorize each app in the browser the first time it comes up. No to create, no gateway to configure.

Outcomes

Install the Arcade plugin in your coding agent and run your first task against a connected app.

Prerequisites

  1. Node.js  so you can run npx
  2. A supported : Cursor, Claude Code, VS Code, GitHub Copilot CLI, or Codex
  3. An Arcade

The plugin is built for personal use. When you are ready to put a workflow in front of a team, read from personal use to team rollout. You will want your own MCP Gateway with a curated list and your own identity provider.

Install

Terminal
npx plugins add ArcadeAI/arcade-plugin

The CLI detects the supported agents on your machine and installs to each one. To install to a single , pass --target:

Terminal
npx plugins add ArcadeAI/arcade-plugin --target cursor
npx plugins add ArcadeAI/arcade-plugin --target claude-code
npx plugins add ArcadeAI/arcade-plugin --target vscode
npx plugins add ArcadeAI/arcade-plugin --target copilot
npx plugins add ArcadeAI/arcade-plugin --target codex

To see what an install would write without changing anything:

Terminal
npx plugins discover ArcadeAI/arcade-plugin

Reload your afterwards if it does not pick up the plugin on its own. A couple of agents gate plugins behind a setting. Claude Code may ask you to enable it with claude plugin, and VS Code needs chat.plugins.enabled set to true.

Two things worth knowing if you use more than one . Codex and ChatGPT share a plugin directory, so one install covers both. And VS Code can discover a copy installed through GitHub Copilot CLI, so install in one place rather than both.

Using Claude Desktop or an that is not listed here? See Claude Desktop or any MCP client.

Tools only

If you want Arcade without the skills, commands, and subagent, Cursor and VS Code can add the gateway on its own:

For any other client, add the gateway URL by hand. See any MCP client.

Sign in

You authorize twice, and the two are doing different jobs.

  1. Into the gateway. Your client asks you to authenticate when it first connects. You sign in to Arcade in the browser, and the client can reach the gateway from then on.
  2. Into each app. When a task needs Gmail, Slack, Linear, or anything else, your hands you a link that grants the scopes that task requires. Approve it once and Arcade holds the token for that app.

You never give your agent an or a password for either layer. Ask your to run /arcade-status to see where you stand, or /arcade-connect google to authorize an app before you need it.

Every install points at the same gateway URL, but your tool calls, tokens, and secrets route through your own Arcade , isolated from every other user of that gateway. No one else’s can reach your apps.

Every Arcade gateway works this way, not just the one in the plugin. Server-level and tool-level authorization goes deeper on the split.

What you get in each agent

Every install connects to the same Arcade gateway and gets the same tools. What differs is how much of the rest of the plugin each can load.

AgentToolsSkillsSubagentCommandsRuleHooks
Cursor✅2✅3✅✅
Claude Code✅2✅3—✅
Claude Cowork✅2✅3—✅
GitHub Copilot CLI✅2✅———
VS Code✅2————
Codex and ChatGPT✅2————
Claude Desktop✅—————
Any other MCP client✅—————

Here is what each of those is for.

Commands are the part you type. /arcade-status reports whether the gateway is reachable, whether you have signed in, and which apps you have authorized. /arcade-connect <app> authorizes an app on demand, so you are not interrupted mid-task. /arcade-apps lists what you have connected and lets you disconnect one.

Skills are instructions your reads on its own, so you never have to invoke them. try-arcade teaches it to turn a request like “reply to that thread” into the right sequence of calls across your apps. scale-arcade handles the other kind of question, the one about moving a workflow onto a team gateway, and it walks you through the decisions on this page against your own workspace.

The subagent, arcade-operator, runs an app task in its own and reports back just the result. Arcade exposes a lot of , and searching them fills a conversation with noise that pushes out your actual work. The operator absorbs that.

The rule and hooks are Cursor and Claude Code specific. They tell your that Arcade exists and when reaching for it is the right move, which is the difference between a that sits unused and one your agent actually picks up.

The columns thin out because the Agent Plugins  1.0 specification only makes skills and servers portable. Commands, subagents, hooks, and editor rules are specific to each , so the plugin ships them where the agents that support them will find them.

Try it

Ask your for an outcome rather than naming a :

  • “What’s on my calendar tomorrow?”
  • “Summarize the unread email from this week.”
  • “Draft a reply to that thread, then wait for me before sending it.”
  • “What can Arcade do?”

Your asks you to confirm before it sends, creates, or deletes anything.

Next steps

Last updated on
About ArcadeClaude Desktop