A human-in-the-loop review layer for AI coding agents. You annotate markdown plans with Fix, Question, or Highlight markers using keyboard shortcuts, and agents read those annotations directly through 27 MCP tools. No copy-paste step. The agent applies fixes and reports back with inline diffs, you approve or reject from CodeLens buttons, and quality gates block progress until your conditions are met. Session handoffs preserve context across conversations. Annotations live as HTML comments in the markdown itself, so they survive git and render anywhere. Reach for this when you want structured plan review before implementation, especially across multi-session AI workflows where context loss is expensive.
claude mcp add --transport stdio yeominux-md-feedback uvx md-feedback