Anthropic's Claude Code has seen a 300% increase in usage over the past three months, but its ease in generating large amounts of code without clear author intent poses challenges for engineering teams. These challenges are exacerbated when AI-generated code results in extensive, complex pull requests (PRs) that are difficult to review and understand, leading to potential integration risks. To address this, the team at Graphite focused on teaching Claude Code to use stacked diffs, breaking down large PRs into smaller, manageable, and reviewable tasks. This approach aligns with engineering best practices, enhancing the clarity and reviewability of the code. The GT MCP framework was developed to help AI agents, including Claude, generate stacked PRs automatically, transforming large diffs into a sequence of smaller, focused pull requests. This framework has allowed Graphite engineers to build features, resolve bugs, and refactor code efficiently, while also contributing positively to the broader developer community by saving time and improving workflow.