SCRUM MCP

When multiple AI agents edit code at the same time, politeness is not enough. They need shared state and hard rules. SCRUM MCP started with three mechanisms:

1. Intent declarations - agents announce what they plan to change before touching any files
2. File claims - atomic locks prevent two agents from editing the same file
3. Evidence receipts - agents must prove their work with command output before releasing claims

The useful bit was not making agents sound clever. It was making them behave around files, evidence and shared context. Later versions added compliance checks, sprint collaboration, role-based tool permissions, budget tracking, audit history and a searchable knowledge base.

The real test was running a group of named agents across several web projects at once. Very quickly the problem stopped being "can the model write code?" and became "can it see the right work, avoid touching the wrong files, and prove what it did?"

The inbox pre-processor solved one of the least glamorous but most important problems: agents could not call the APIs they were being told to use. So the system brought the task queue to them instead. Result: productivity went from zero completions to 12 issues resolved in 2 hours.