MCP Tooling

The gap between official and capable.

In practice a connection can mean bringing the relevant records into an investigation, preparing a request for someone to review, or carrying an approved change into the system where the work happens. The Model Context Protocol (MCP) is the standard interface that makes this possible: it translates what an AI model wants to do into calls to a platform’s API. Most platforms that publish official MCP servers expose a curated subset of that API, covering the most common use cases. For general workflows that is often sufficient. For production automation it usually isn’t.

Take Google Workspace. The Gmail, Drive, and Calendar connectors that ship with the major AI assistants cover the everyday cases well: search your mail, open a file, see your week. The Google Workspace APIs go much further. They reach a document's structure and revision history, a spreadsheet by range, a slide by element, a file's permissions and the text inside a scanned PDF, and, through the Admin APIs, the users, groups, and devices of a whole organization.

We built our own Google Workspace connection on those APIs. A Team Member or an agent can now file an attachment straight into the right Drive folder, make a structured edit to a document, read a file too large for a chat window, or, where a client has authorized it, work across that client's accounts under the admin controls Google provides. Each tool does one specific thing, under the permissions of the person or agent calling it, and every call is logged against them.

The same principle applies to any platform with a documented API. If it can be called programmatically, it can become a tool.