AI coding agents have no shared workspace across a microservice stack so cross-repo refactors mean running four blind sessions at once

Coding agents running on a microservice stack have no shared workspace. When a refactor touches three repositories, each agent session starts blind to what the others are doing, reads stale dependency contracts, and opens PRs that conflict at merge time. The result is a distributed systems problem inside the agent layer itself: no shared state, no conflict detection, no atomic commit across repo boundaries. This MCP server gives every coding agent in the stack a live shared context index, a cross-repo dependency graph, and a coordinated change ledger so a multi-repo refactor runs as a single coherent operation rather than several blind parallel sessions.