Skip to main content

Automation

Some work is routine: the same inbox checked every morning, the same messages sorted, the same replies drafted. An AI agent can do that work — but only if it can reach your website, your Slack, and your mail. Bloque is how it reaches them: one endpoint, one API key, every MCP server your admin has already set up.

What an automation needs​

An unattended agent has no one to click "approve" for it, so the setup has to be simple:

  • One credential. The script holds a single Bloque API key, not a separate token for each service. Nothing else has to be stored on the machine that runs it.
  • Tools it can reach. Every MCP server configured in your Hub is available through the same connection. Add a server in Bloque and the automation can use it on its next run — no redeploy, no config change.
  • A record of what it did. Every tool call appears on the Logs page, so you can check what the agent touched while no one was watching.

Example: triaging contact-form inquiries​

The inquiry-triage sample project is a complete, working automation you can copy and adapt.

It handles the contact form on a WordPress site. On each run it:

  1. Fetches the newest unprocessed submissions from WordPress (Contact Form 7 + Flamingo)
  2. Reads each one and decides what it is — a genuine business inquiry, a job application, a sales pitch, or noise
  3. Acts on it: posts real inquiries to a Slack channel, drafts a reply email for job applicants, DMs the admin anything it isn't sure about
  4. Marks the submission processed so it is never handled twice

Three MCP servers are involved — WordPress, Slack, and Zoho Mail — and the project itself knows about none of them. They live in a Hub, and the whole .mcp.json is this:

{
  "mcpServers": {
    "bloque": {
      "type": "http",
      "url": "https://mcp.bloque.run/mcp",
      "headers": {
        "Authorization": "Bearer ${BLOQUE_API_KEY}"
      }
    }
  }
}

The judgment calls — which inquiries matter, what counts as too old to act on, when to ask a human — live in a plain-English Claude Code skill file in the project, so you can change the rules by editing a Markdown file.

Two of the servers it uses have setup guides here: WordPress MCP and Create a Slack App.

Connecting your own script​

Anything that speaks the MCP HTTP transport can connect the same way — a Claude Code project, an MCP SDK script, or a workflow tool such as n8n or Dify:

URL:           https://mcp.bloque.run/mcp
Authorization: Bearer <your API key>

Create the key under API Keys in the sidebar, and keep it in an environment variable rather than in the file you commit. See Connection Reference for client-specific settings.

Running it on a schedule​

Once it runs correctly by hand, put it on a timer:

  • cron on a machine that is always on — the simplest option
  • GitHub Actions — Claude Code can run on a schedule without a server of your own

Give a scheduled automation its own API key. If it ever misbehaves, you revoke that one key and everything else keeps working.

What's next​

  • Team Rollout — set up Bloque for a team, so people and automations share the same servers
  • Logs — check what an unattended run actually did