Connect your agent.

Grok Bot, or anything that runs a command. One open-source CLI, one skill, the same three permissions.

Jump to: Grok Bot · Any other agent · Troubleshooting

Start with read and draft.

You pick permissions on the Allow screen in SuppyHQ. suppyhq auth login ticks the first two. Leave sending off until you trust its drafts.

Read conversations + customer profiles

Threads, customers, plans, purchases. Nothing changes.
Write drafts

Saves a reply in your composer. Nobody gets it until you hit send.
Send replies to your customers

Off unless you tick it. Every send waits 30 seconds so you can cancel, and the email names the agent.

Remove an agent any time from Integrations → Agents in the app.

Letting it send.

You can let your agent email customers on its own, at your own risk. Run suppyhq auth login --allow-send and tick "Send replies to your customers" on the Allow screen. The rails stay on: a reply to someone you haven't written to yet still waits for you as a draft, every send waits 30 seconds so you can cancel it, and the email names the agent that wrote it. It still goes out from your inbox, so you're responsible for what it says.

Grok Bot

Your Bots work on their own cloud computer with a terminal and a browser. The CLI goes there, not on your laptop.

Paste this to your Bot.
When it hands you the computer, sign in to SuppyHQ in the Bot's browser and click Allow. Leave "Send replies" unticked.
Ask it something real: "What's open in SuppyHQ that can't wait?" or "Draft a reply to the refund thread."

Message to your Bot

Connect to my SuppyHQ inbox.
1. Install the SuppyHQ CLI on your computer: curl -fsSL https://suppyhq.com/install-cli | bash
2. Run `suppyhq auth login --name="Grok Bot"` in the background. It waits up to 5 minutes.
3. Open the link it prints in your browser and hand me the computer. I'll sign in and click Allow.
4. Run `suppyhq doctor` and tell me what it says.
5. Save ~/.claude/skills/suppyhq/SKILL.md as a skill called SuppyHQ.
Never send a reply unless I say "send". Draft instead.

Click Allow in the Bot's browser, not yours. The approval returns to the CLI on the Bot's computer.
Every Bot on your account shares that computer, so they all share this login.
If the skill doesn't show up when you type /, enable it for the Bot under Settings → Plugins → Yours.
Why not a plugin? Grok Bot plugins connect to a public MCP URL. SuppyHQ's MCP server runs on the same computer as the agent. The CLI does the same reads and drafts.

Any other agent

It runs shell commands? Install and log in on the computer it uses, then point it at ~/.claude/skills/suppyhq/SKILL.md. Every command returns JSON.

Terminal

curl -fsSL https://suppyhq.com/install-cli | bash
suppyhq auth login
suppyhq doctor

It speaks MCP? Register suppyhq mcp as a local (stdio) server after logging in. Most clients take this shape. Use ["mcp", "--read-only"] to hide the reply tool.

MCP config

{
  "mcpServers": {
    "suppyhq": {
      "type": "stdio",
      "command": "suppyhq",
      "args": ["mcp"]
    }
  }
}

Three tools: suppyhq_conversations, suppyhq_customers, suppyhq_identity. Claude Code, Cursor, Codex, Copilot CLI, Gemini CLI and opencode have one-line installs on the agents page.

Troubleshooting

Start here: suppyhq doctor checks the CLI, the login and the skill, and tells you the fix.
suppyhq: command not found: The installer puts it in ~/.local/bin and prints the line that adds it to your PATH. Run that, or call ~/.local/bin/suppyhq directly.
I clicked Allow and nothing happened: Allow hands the login back to the CLI at 127.0.0.1, so it has to happen in a browser on the same computer, within 5 minutes. Run suppyhq auth login again and approve there.
not authenticated from suppyhq mcp: The MCP server uses the CLI's login. Run suppyhq auth login first, then restart the agent.
403 Forbidden on a reply: The agent doesn't have that permission. Nothing was sent. Without send, it should save a draft instead. To allow sending, run suppyhq auth login --allow-send.
The agent doesn't know the skill: Restart the session. Check the file exists at ~/.claude/skills/suppyhq/SKILL.md. If it doesn't, run suppyhq install-skill.

The CLI, skill and MCP server are open source (MIT): github.com/karloscodes/suppyhq-cli. Read next: AI agents · Security