Connect your AI

Your plgn address

Every client needs the same one thing. You will find it in your dashboard under Settings → Connect your AI, with a button to copy it.

https://useplgn.com/api/mcp

Different products call the box different things: a custom connector, a remote MCP server, an integration URL, a context server. They all mean this address.

1. 1Paste the addressInto whatever your client calls a connector or an MCP server. Give it the name plgn so you recognise it later.
2. 2Approve it once, in your browserThe first time the client reaches plgn, your browser opens and asks you to approve access. Approve it and you are done. The connection is remembered.
3. 3Ask it to do something“What brands are in my plgn workspace?” is the fastest way to prove it worked. If it answers, you are connected.

Pick your assistant

Every one of these can reach a plgn workspace. Choose yours and it takes you to its instructions.

Claude CodeThe full experience, 27 commandsClaude DesktopMac and Windows appClaude on the webclaude.ai in a browserChatGPTConnectors, with one limitCodex CLIOpenAI, in a terminalGemini CLIGoogle, in a terminalopencodeOpen source, in a terminalCursorOne JSON file and a bridgeVS CodeWith GitHub CopilotWindsurfNow Devin Desktop, with a bridgeClineRemote servers panelZedContext serversWarpThe terminalGooseRemote extensionLibreChatSelf-hosted, one YAML blockJetBrains IDEsAI Assistant

Not listed? See anything else that speaks MCP. The steps are the same everywhere.

Claude Code

The only client that gets the whole product. Install the plugin and you get twenty-seven /plgn commands, the rulebooks behind them, and the workers that write a month in parallel.

/plugin marketplace add PLGN-App/PLGN-CLAUDE
/plugin install plgn

Then restart Claude Code and run /plgn setup. The install page covers this properly, including the one step everybody skips.

Claude Desktop and Claude on the web

One set of steps for both, because it is one setting. A connector added in the app appears on claude.ai, and one added on claude.ai appears in the app: you do this once per Claude account, not once per device.

1. 1Open SettingsIn the app it is the Claude menu on Mac and the app menu on Windows. On the web it is under your name, bottom left.
2. 2Go to ConnectorsThen Add custom connector.
3. 3Paste the addressName it plgn, paste the address, and save.
4. 4Approve in your browserClaude opens a browser window once. Approve it and the tools appear.

ChatGPT

In ChatGPT, add a custom MCP server. OpenAI has called it a connector, an app and a plugin, and the menu moves with the name, but the steps stay the same: paste the address, choose OAuth if it asks how to sign in, and approve access in the browser window it opens.

Custom MCP servers need a paid ChatGPT plan, and developer mode may have to be turned on first; in a Business or Enterprise workspace, an admin does that. If your plan does not offer them, nothing in plgn can change that. Claude Desktop is free to install and connects to the same workspace.

Command-line assistants

Every major terminal assistant now reads MCP servers from a config file. Three are set out below, and the rest follow the same shape with the same address.

Codex CLI

OpenAI's terminal assistant keeps its servers in ~/.codex/config.toml.

[mcp_servers.plgn]
url = "https://useplgn.com/api/mcp"

Older versions of Codex only spoke to servers they could launch as a command. If yours will not take a url, update it: remote servers arrived later than the file did.

Gemini CLI

Google's terminal assistant uses ~/.gemini/settings.json, and names the field httpUrl for a streaming HTTP server.

{
  "mcpServers": {
    "plgn": {
      "httpUrl": "https://useplgn.com/api/mcp"
    }
  }
}

Then run /mcp inside Gemini CLI to see plgn and its tools listed.

opencode

opencode keeps its configuration in opencode.json, in your project or in ~/.config/opencode/.

{
  "mcp": {
    "plgn": {
      "type": "remote",
      "url": "https://useplgn.com/api/mcp",
      "enabled": true
    }
  }
}

Everything else with a terminal

Amazon Q Developer CLI, Continue, Crush, Aider and the rest all take a remote MCP server one way or another, and they all take the same address. Look in their configuration for mcpServers, mcp_servers or an Add MCP server menu, and give it the URL rather than a command. The checklist further down is what any of them will ask for.

Cursor

Cursor connects through mcp-remote, a small open-source bridge that runs on your computer. Cursor's own sign-in goes back to the app through a cursor:// link, and plgn does not accept those: any app on a computer can claim such a link, so plgn cannot know where the sign-in would end up. The bridge signs in through your own computer instead. It needs Node.js 18 or newer.

{
  "mcpServers": {
    "plgn": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://useplgn.com/api/mcp"]
    }
  }
}

~/.cursor/mcp.json makes plgn available in every project; .cursor/mcp.json inside a project limits it to that one. Reopen Cursor, then check Settings → MCP: plgn should be listed with its tools. The first time the bridge starts, it opens your browser to approve access.

VS Code

VS Code reaches MCP servers through GitHub Copilot's agent mode. Put this in .vscode/mcp.json in your project, or add it to your user settings to have it everywhere.

{
  "servers": {
    "plgn": {
      "type": "http",
      "url": "https://useplgn.com/api/mcp"
    }
  }
}

Open the Chat view, switch it to Agent, and plgn's tools appear in the tools list. The first time it connects, VS Code asks you to let plgn sign in, then opens your browser to approve access. Copilot asks before running each tool the first time.

Windsurf

Windsurf, now called Devin Desktop, connects through the same mcp-remote bridge as Cursor, which signs in through your own computer and needs Node.js 18 or newer. Its MCP servers live in a file called mcp_config.json, which has moved between versions, so open it from the editor: in the Cascade panel, the … menu has Open MCP config file.

{
  "mcpServers": {
    "plgn": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://useplgn.com/api/mcp"]
    }
  }
}

Save the file and restart the editor. The first time the bridge starts, it opens your browser to approve access.

Cline

1. 1Open the MCP Servers panelThe icon is in Cline's top bar.
2. 2Choose Remote ServersNot Installed, and not the marketplace, plgn lives at an address.
3. 3Add itName plgn, paste the address, save.

Cline shows the server as connected once it has read the tool list, and opens your browser for approval the first time it calls one.

Zed

Zed calls them context servers. Open the assistant panel's settings, add a custom context server, and give it the address.

Zed's settings file moves between versions, so add it through the panel rather than by hand, the editor writes the right shape for the version you have.

Warp

Warp is a terminal with an assistant in it, and it takes MCP servers under Settings → AI → MCP servers. Add one, choose the remote or URL option, and paste the address.

Worth knowing if you already work in a terminal: Claude Code runs inside Warp perfectly well, and the plugin gives you far more than the bare tools do.

Goose

Run goose configure, choose Add Extension, then Remote Extension, the streaming HTTP kind, not the command kind, and give it the address.

What type of extension?   Remote Extension (Streaming HTTP)
Extension name            plgn
Endpoint URL              https://useplgn.com/api/mcp

LibreChat

LibreChat is self-hosted, so this goes in your librechat.yaml and takes effect when you restart it.

mcpServers:
  plgn:
    type: streamable-http
    url: https://useplgn.com/api/mcp

Because LibreChat runs on your own server, the browser approval happens on the machine you open LibreChat from, exactly as it does anywhere else.

JetBrains IDEs

IntelliJ, PyCharm, WebStorm and the rest share one setting: Settings → Tools → AI Assistant → MCP. Add a server, choose the URL option, paste the address.

The AI Assistant plugin has to be installed and signed in first: plgn appears in its tool list, not in the IDE's own menus.

Anything else that speaks MCP

The list above is not a limit. plgn is an ordinary remote MCP server over HTTP with OAuth sign-in, so anything that supports that standard can use it, including one you write yourself.

What you get in each

Two things are worth being clear about before you choose, because they are easy to expect and not get.

In plain terms: anywhere else you can say “write me five posts about the new pricing page and save them as drafts”, and it will. What you do not get is the plan-then-one-confirmation shape, the rules that stop a month bunching into one week, or the ability to take a whole run back in one step.

Can the plugin itself be used anywhere else?

No, and it is worth being exact about why. A plugin is a Claude Code format: the /plgn commands, the rulebooks and the workers are files Claude Code knows how to read, and no other client has anything that reads them. There is no version of it for ChatGPT, Cursor or a terminal assistant, and there will not be one until those products have a plugin format of their own.

What every other client gets instead is the tools, which is most of the value. Ask in plain language: “write five posts about the new pricing page, in our voice, and save them as drafts”. Your assistant reads the brand and writes them. What it does not have is the plan-then-one-confirmation shape, the rules that stop a month bunching into one week, and the marker that makes a whole run undoable.

When it does not connect