A Gmail connector was sending an AI model more email data than users needed. Rebuilding it reduced token use by 30 to 75 percent, depending on the workflow, according to a LinkedIn post by Workato CIO Carter Busse.
Busse also reported 41 percent faster performance. He said Josh Cramoy completed the rebuild in one day using Codex and Workato AIRO MCP Server.
Why the connector needed work
Internal feedback flagged excessive token use in the Gmail MCP connector supplied with an unnamed large language model.
Cramoy benchmarked the connector and inspected its responses. According to Busse, those responses included full email headers and MIME details, technical information about how messages and attachments are structured.
That information can serve a purpose, but it was more than users needed for the workflows described.
MCP, short for Model Context Protocol, lets AI applications connect to external tools and data. What those tools return affects how much information a model must process.
The reported gains suggest that companies should inspect connector responses when investigating slow or expensive AI workflows. Unnecessary email data can add processing overhead before a model answers the user.
Lower token use may reduce costs where pricing depends on token consumption. Busse did not provide dollar savings.
For teams building AI email tools, the practical lesson is specific: check what the connector sends back and whether the task actually requires it.