Connect an MCP Client

Connect Claude, Cursor or any MCP client to your inbox with an API key

Connect an AI agent to your inbox in three steps: create an API key, copy the server URL, and paste one configuration into your MCP client. The agent works with the same contacts and conversations as your team, limited to what its key allows.

Before You Start

  • Plan. MCP needs the Pro or Enterprise plan. On a lower plan the MCP page shows an upgrade prompt instead of the setup.
  • Role. Owners and Admins create the API key the agent uses.
  • Where. Open Settings, then Webhooks and API, then MCP.

Connect a Client

  1. Create a key just for the agent. On the MCP page, click Manage API keys under Authentication, then follow API Keys in the Dashboard. Tick only the permissions the agent needs.
  2. Under Server URL, click Copy. The server speaks streamable HTTP, so add it as a remote MCP server in your client.
  3. Under Client configuration, click Copy and paste the configuration into your client's MCP settings.
  4. Replace YOUR_API_KEY with the key from step 1.

The configuration looks like this, with your own server URL:

{
  "mcpServers": {
    "wexio": {
      "url": "<the Server URL from the page>",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

The key goes in the Authorization header as a Bearer token. The key's permissions decide which tools the agent can use, and each tool checks them on every call.

What the Agent Can Do

The What the agent can do section lists the tool groups. Each is available only when the key holds the matching permissions.

Group in the dashboardWhat it coversTool pages
ContactsFind, read and update contactsContacts
ConversationsRead conversations, assign and close themInbox reads, Chat control
MessagingSend replies and templates on your channelsMessaging
CollectionsSort conversations into your collectionsInbox reads
ChannelsSee which channels are connectedChannels
TeamLook up members to assign work toTeam

The server also has knowledge document tools. Every tool and the scope it needs is listed in the MCP overview.

Give an agent a reply-only key: Read messages and Send messages, without Start new conversations. If someone talks the model into misbehaving, it still cannot message a contact first. See Connecting an AI Agent.

Check the Agent Is Refused What It Should Be

Open API keys, then click Logs on the agent's key. Refused calls show the reason, for example Missing permission. Revoking the key cuts the agent off at once, even in the middle of a session.

Was this page helpful?

On this page