Overture MCP

Content

Overture MCP is a Model Context Protocol server for visualizing and approving AI coding-agent plans before execution. It lets an agent submit plan XML, opens a local interactive graph UI, waits for human approval, tracks node status, handles branches, supports plan updates, and records plan history for later review.

It is most useful when the user wants a structured checkpoint between "the agent has a plan" and "the agent starts editing." Overture can make the plan more inspectable, but the actual safety still depends on approvals, tests, review, and the permissions of the coding agent's other tools.

Source Review

• https://github.com/SixHq/Overture
• https://github.com/SixHq/Overture/blob/main/README.md
• https://registry.npmjs.org/overture-mcp
• https://github.com/SixHq/Overture/blob/main/packages/mcp-server/package.json
• https://github.com/SixHq/Overture/blob/main/packages/mcp-server/src/index.ts
• https://github.com/SixHq/Overture/blob/main/packages/mcp-server/src/http/server.ts
• https://github.com/SixHq/Overture/blob/main/packages/mcp-server/src/storage/project-storage.ts
• https://github.com/SixHq/Overture/blob/main/LICENSE

These sources were reviewed on 2026-06-06. Prefer the live repository, README, npm registry metadata, MCP package metadata, server source, local HTTP server source, storage source, and license file for current install commands, tool behavior, UI behavior, storage paths, and licensing.

Features

• npm package overture-mcp.
• Stdio MCP server launched with npx -y overture-mcp.
• Local visual plan canvas with plan nodes, branches, complexity, risks, and status.
• Approval workflow through submit_plan and get_approval.
• Execution tracking tools for node status, plan completion, failures, pause, rerun, resume information, and plan updates.
• Node detail update tools for titles, descriptions, expected outputs, and risks.
• Local project history and global fallback history storage.
• Local UI, WebSocket updates, attachment saving, and file viewing endpoints.
• MIT license.

Installation

Configure the MCP server in your client:

{
  "mcpServers": {
    "overture": {
      "command": "npx",
      "args": ["-y", "overture-mcp"]
    }
  }
}

After restarting the MCP client, ask Claude to create a plan in Overture before editing, then review the local plan graph and approve only the steps and branch you want executed.

Use Cases

• Review an AI coding agent's plan before it edits files.
• Compare branches in a proposed implementation approach.
• Attach context files or meta-instructions to specific plan nodes.
• Pause, resume, rerun, or mark execution progress during long tasks.
• Persist plan history for a project.
• Track which plan nodes produced outputs, failures, or follow-up work.
• Add a human approval checkpoint to high-risk coding workflows.

Safety and Privacy

Overture improves visibility into an agent plan, but it does not sandbox the agent or guarantee that approved steps are correct. Keep normal safeguards in place: scoped tool permissions, tests, code review, and careful handling of destructive operations.

Plan data can contain private task details, workspace paths, attached files, secret fields, node outputs, and execution history. Overture stores history and attachments locally and exposes them through a local UI. Use it only with trusted local clients, avoid attaching secrets unless needed, and review project-level Overture files before committing or sharing a repository.

Duplicate Check

Existing MCP content includes code assistants, review tools, and workflow helpers, but no dedicated entry for SixHq/Overture or the overture-mcp npm package was found in content/mcp. This entry is distinct because it focuses on visual plan approval, branch selection, execution tracking, and plan history for AI coding agents.