What if your AI coding agent had to show you its plan before writing a single line of code?
That is the pitch behind Overture, an open-source MCP server (Model Context Protocol - the standard for connecting AI tools to external services) that intercepts an AI agent's execution plan and renders it as an interactive visual flowchart. Instead of watching Claude Code or Cursor barrel through file changes and hoping for the best, you get a drag-and-drop canvas showing every step, dependency, and branch point before implementation starts.
How It Works
Overture runs locally and plugs into Claude Code, Cursor, Cline, and GitHub Copilot (VS Code 1.99+). Installation for Claude Code is a one-liner: claude mcp add overture-mcp -- npx overture-mcp. For other editors, you add a JSON config block.
Once connected, the agent's plan streams into a flowchart in real time. Each node shows a description, complexity rating, expected output, risks, and edge cases. You can:
- Approve, reject, or modify individual steps before they run
- Add custom instructions to specific nodes ("use PostgreSQL here, not SQLite")
- Compare branch options side-by-side with pros, cons, and complexity ratings
- Pause execution mid-plan with the spacebar and resume when ready
- Re-run failed nodes or restart from any point in the plan
The structured output panel tracks files changed, created, and deleted, plus packages installed and tool calls made.
Practical Value
The 593 GitHub stars and active development (81 commits, MIT license) suggest this is filling a real gap. Current AI coding tools are fast but opaque - they make architectural decisions without explaining them, install dependencies you did not ask for, and restructure files in ways that are hard to undo. Overture adds a review layer without slowing down the parts of AI coding that actually work well.
Plan history auto-saves every 3 seconds locally, so you can revisit and compare previous approaches. Multi-project support with tabs makes it usable across different codebases in the same session.