MCP server: connect an AI assistant

WbizTool's MCP server lets an AI assistant work with your workspace. You can ask it things like "which messages failed today, and why?", "pause the rent reminder" or "send our new opening hours to these five customers", and it uses your WbizTool account to do them.

It speaks the Model Context Protocol (MCP), which Claude Code, Cursor, VS Code and other AI tools understand. Behind the scenes every action is a call to API v2 made with your own API v2 key, so the key decides what the assistant may see and do.

What an assistant can do#

The assistant only sees the tools the key allows (see scopes):

It also offers three ready-made prompts: daily summary, failed messages and draft an announcement.

It can't connect a new number (that stays in your dashboard), upload files, manage webhooks or read your Unibox inbox.

1. Make a key for the assistant#

1. Open the API v2 keys card

   In WbizTool, go to Settings → API keys and find the API v2 keys card.

2. Use the AI assistant preset

   Click Create v2 key, then For an AI assistant. It makes a read-only key that can read messages and numbers and expires in 90 days – a safe way to start: the assistant can look, but can't send or change anything.

3. Allow sending only if you want it

   If the assistant should send messages or manage reminders, untick Read-only and tick Send messages (and Reminders if needed). You can also limit the key to some of your numbers.

4. Copy the key

   The key (wbz_live_…) is shown once. Keep it in an environment variable on your computer, such as WBIZTOOL_API_KEY – never paste it into a file you share or commit.

export WBIZTOOL_API_KEY=wbz_live_…

2. Add the server to your AI tool#

claude mcp add --transport http wbiztool https://mcp.wbiztool.com/mcp \
  --header "Authorization: Bearer ${WBIZTOOL_API_KEY}"

{
  "mcpServers": {
    "wbiztool": {
      "type": "http",
      "url": "https://mcp.wbiztool.com/mcp",
      "headers": { "Authorization": "Bearer ${WBIZTOOL_API_KEY}" }
    }
  }
}

{
  "mcpServers": {
    "wbiztool": {
      "url": "https://mcp.wbiztool.com/mcp",
      "headers": { "Authorization": "Bearer ${env:WBIZTOOL_API_KEY}" }
    }
  }
}

{
  "inputs": [
    { "type": "promptString", "id": "wbiztool-key", "description": "WbizTool API v2 key (wbz_live_…)", "password": true }
  ],
  "servers": {
    "wbiztool": {
      "type": "http",
      "url": "https://mcp.wbiztool.com/mcp",
      "headers": { "Authorization": "Bearer ${input:wbiztool-key}" }
    }
  }
}

• Claude Code: run the command once (add --scope project to share it with a project through .mcp.json; the file holds ${WBIZTOOL_API_KEY}, not the key).
• Cursor: put it in ~/.cursor/mcp.json, or .cursor/mcp.json in a project.
• VS Code: put it in .vscode/mcp.json. VS Code asks for the key once and keeps it in its secret storage.

Then ask the assistant something like "What's my WbizTool account?" – it should answer with your workspace's name and plan.

3. How sending stays safe#

• Every send is previewed first. When you ask for a message, the assistant first gets a preview from WbizTool – which number it goes from, to whom, the exact text, any file, when – and nothing has been sent yet. It's told to show you that preview and wait for your yes. Only then does it send, with exactly what you saw.
• A retry never sends twice. If the assistant repeats a confirmed send (for example after a timeout), WbizTool answers with the message it already sent.
• The key is the hard limit. A read-only key can't send at all. A key without Send messages never even sees the send tools, and a key limited to some numbers can't use the others. WbizTool's own limits apply too: 60 requests a minute per key and at most 1,000 messages a day per number.
• Text from other people is marked as data. Messages, names and files that came from outside reach the assistant marked as untrusted, and it's told never to follow instructions inside them.
• No secrets in answers. QR codes and pairing codes are never shown, and lists don't include file links.
• Revoke at any time. Revoking the key on Settings → API keys stops the assistant at once.

Privacy#

When you connect an assistant, what it reads from WbizTool (messages, names and numbers you let the key see) is sent to that assistant's provider – for example Anthropic for Claude, or OpenAI for ChatGPT – and handled under their terms. WbizTool's MCP server itself keeps nothing: it passes each request to API v2 and logs which tool was used, never message text or your key. Give the key only the access the assistant needs.

Troubleshooting#

Related#

• API v2 overview
• API keys
• Messages API v2
• Apps & integrations