🌉 Godot MCP Bridge

Close the AI self-correction loop 🎯 — your AI agent writes code, runs the game, sees real runtime feedback, and fixes its own mistakes. Autonomously.

Godot MCP Bridge gives your AI agent eyes and hands inside the engine: live errors, the scene tree, node properties, screenshots, even interactive breakpoints — all fed straight back to the agent, so it can verify its own work with real feedback instead of assumptions.

✨ What you get (40+ tools)

• 🔴 Runtime feedback — live script errors, print() output, scene tree, node inspection, screenshots
• 🐞 Interactive debugging — breakpoints, call stack, frame variables, step into/over/out, evaluate expressions in paused frames
• 🎮 Act on the engine — set properties, call methods, evaluate expressions, simulate input, play/stop the game
• 🖼️ Editor assets — list project assets, view images (Vision-ready), slice sprite sheets, inspect SpriteFrames, audio waveforms, reverse dependency lookup
• ⚡ Editor efficiency — refresh filesystem, reload scene/resource/plugin, GDScript diagnostics (syntax + type + warnings via LSP)

Works with any MCP-compatible client: Claude Code, Codex, DeepSeek Harness, and more.

📦 Two parts

1. This addon — the editor-side bridge (installed from this page).
2. @letsagents/godot-mcp — the MCP server your AI client spawns (installed from npm).

The addon alone is not enough; both parts are required.

🚀 Install

✅ Requirements

• Godot 4.7+
• Node.js >= 20
• An MCP-compatible AI client

1️⃣ Editor addon

Install from the Asset Store (keep only the addons/ folder), then enable it:

Project Settings → Plugins → Godot MCP Bridge → Enable

2️⃣ MCP server

node --version                 # must print v20 or higher
npm install -g @letsagents/godot-mcp

💡 Can't install globally? In the configs below, set command to "node" and add the args ["/absolute/path/to/mcp-server/server.js"].

3️⃣ AI client — pick yours

Claude Code — .mcp.json in your project root:

{
  "mcpServers": {
    "godot-mcp": { "command": "godot-mcp" }
  }
}

Codex — ~/.codex/config.toml:

[mcp_servers.godot-mcp]
command = "godot-mcp"

DeepSeek Harness — add to the cordis.yml composition file in your workspace (the directory where you run dsh):

- id: mcp-godot
  name: '@deepseek-ai/dsh-mcp-client'
  config:
    serverName: godot
    transport: stdio
    command: godot-mcp

Open your project in the Godot editor — the bridge listens on 127.0.0.1:6510 automatically and your AI client connects to it. The loop begins. 🔁

🧩 Multiple editors at once

One bridge = one port = one client. The default setup above needs zero config. To run two Godot editors side by side (e.g. two projects), give each project its own port and point each client at the matching one:

Project B — project.godot:

[godot_mcp]
bridge_port=6511

Project B's client — set GODOT_MCP_BRIDGE_PORT accordingly:

{ "command": "godot-mcp", "env": { "GODOT_MCP_BRIDGE_PORT": "6511" } }

env = { GODOT_MCP_BRIDGE_PORT = "6511" }

    env:
      GODOT_MCP_BRIDGE_PORT: '6511'

(Project A keeps the default 6510 — nothing to change.)

📝 Notes

• The bridge is local-only (127.0.0.1, default port 6510, configurable) and serves one client at a time.
• C# projects: engine-level feedback (scene tree, errors, screenshots, …) works regardless of language; script breakpoints apply to GDScript only.

⚖️ License

MIT. Not affiliated with or endorsed by Godot Foundation.