Cocos Creator MCP Servers Compared: Funplay Cocos MCP vs Alternatives

Funplay Cocos MCP is a Cocos Creator editor extension that exposes 67 tools through an MCP server on 127.0.0.1:8765. It gives AI coding agents direct control over scene editing, asset management, script execution, and runtime interaction — without adding runtime dependencies to shipped builds.

This article compares Funplay Cocos MCP against its siblings in the Funplay family — the Unity MCP (79 tools) and Godot MCP (105 tools) — and helps you decide when each one fits.

What Funplay Cocos MCP Actually Is

The package, funplay-cocos-mcp, is a Cocos Creator extension maintained by FunplayAI. Version 0.1.1, released 2026-04-16, ships under the MIT license. It requires Cocos Creator 3.8 or later and runs entirely inside the editor. Nothing from this extension ends up in your game's build output.

Three facts that define its value:

1. 67 registered tools cover script execution, project and scene management, node editing, assets and prefabs, components, UI, camera, animation, files, diagnostics, runtime control, interaction and input, plus screenshots and windows.
2. A core profile of 19 tools lets you expose only high-signal tools, cutting noise in the AI client's tool list.
3. One-click configuration generates ready-to-paste config for Claude Code, Claude Desktop, Cursor, VS Code, Codex CLI, Trae, and Kiro — unique to the Cocos variant, Trae and Kiro support is exclusive.

How It Compares: The Funplay MCP Family

The Godot variant leads in raw tool count at 105. The Unity variant sits in the middle at 79. Cocos comes in at 67, but compensates with the execute_javascript tool — a single entry point that replaces many narrow, single-purpose tools.

The execute_javascript Advantage

Cocos Creator's editor runs on Electron. That means JavaScript is a first-class citizen in both the scene runtime and the editor's Node.js process. Funplay Cocos MCP exploits this with one tool:

// Running in "scene" context — accesses the active scene's runtime
const node = cc.find("Canvas/Player");
node.setPosition(100, 200);

// Running in "editor" context — full Node.js access
const fs = require("fs");
const files = fs.readdirSync(Editor.Project.path + "/assets/scripts");
console.log(files);

The execute_javascript tool accepts a context parameter: "scene" or "editor". In editor context, scripts receive injected globals: Editor, fs, path, os, require, plus helper functions getStatus(), listTools(), readResource(uri), callTool(name, args), and configureClient(targetId). Two aliases — execute_scene_script and execute_editor_script — exist as convenience wrappers.

In Unity and Godot MCPs, achieving the same breadth of control typically requires calling multiple specialized tools in sequence. The Cocos variant's unified script execution model is a direct benefit of Cocos Creator's JavaScript foundation.

Installation Walkthrough

There are two installation paths. The simplest is cloning directly into your project.

Option A: Project-Level Extension

# From your Cocos Creator project root
cd extensions
git clone https://github.com/FunplayAI/funplay-cocos-mcp.git

Then open Cocos Creator, go to Extension → Extension Manager → Refresh. The extension appears in the list.

Option B: Global Extension

Place the cloned folder at ~/.CocosCreator/extensions/funplay-cocos-mcp/. This makes the MCP server available across all projects.

Starting the Server

After installation, a new MCP Server panel appears in the editor. Click Start Server. The HTTP endpoint boots on 127.0.0.1 port 8765.

From that panel, select your AI client — Claude Code, Claude Desktop, Cursor, VS Code, Codex CLI, Trae, or Kiro — and the extension writes the correct configuration to the right file location. No manual JSON editing required.

Here is what a generated Claude Desktop config looks like:

{
  "mcpServers": {
    "cocos-creator": {
      "command": "node",
      "args": ["/path/to/funplay-cocos-mcp/server.js"],
      "env": {}
    }
  }
}

Built-In Prompts and Resources

Beyond tools, the extension ships with 4 built-in prompts and 8 built-in MCP resources.

The prompts are:

• fix_script_errors — Diagnose and repair script compilation errors.
• create_playable_prototype — Scaffold a playable scene from a description.
• scene_validation — Check the active scene for common problems.
• auto_wire_scene — Automatically connect node references and component bindings.

Resources expose editor state — scene hierarchy, selected node properties, asset metadata — that AI agents can read without calling tools.

When to Use Funplay Cocos MCP

Use it when:

• Your team builds in Cocos Creator 3.8 or later.
• You want AI agents to iterate on scenes, prefabs, and scripts without leaving the conversation.
• You value a smaller, composable tool surface (core profile: 19 tools) over sheer quantity.
• You use Trae or Kiro and want zero-config setup.

Use the Unity or Godot variants when:

• Your project runs on Unity 2022.3+ or Godot 4.2+ instead.
• You need more granular tools (79 for Unity, 105 for Godot) rather than script-level access.
• You are working with C# or GDScript ecosystems where per-tool APIs map more naturally.

Gotchas and Limitations

Version lock-in. The extension requires Cocos Creator 3.8 or later. Projects on 3.7 or earlier cannot use it. There is no backwards-compatibility layer.

Editor-only, not runtime. All 67 tools operate inside the editor. You cannot use this MCP server to control a running build on a device. The runtime tools (4 of 67) interact with the editor's play-in-editor simulation, not production builds.

Tool-list noise at scale. The full profile exposes 67 tools. Some AI clients struggle with large tool lists, hallucinating tool names or skipping relevant ones. Switch to the core profile (19 tools) when you notice degraded agent performance. The core profile is a deliberate design choice to mitigate this problem.

No multiplayer or device testing. The interaction and input tools (8 total) simulate taps, drags, and key events in the editor viewport. They do not replace on-device testing for touch accuracy, network latency, or platform-specific input quirks.

JavaScript required. The execute_javascript tool demands familiarity with Cocos Creator's internal APIs. TypeScript users need to understand the JavaScript interop. If your team writes exclusively in TypeScript and never touches editor scripting, the learning curve is real.

Funplay Skills Across the Family

All three Funplay MCP variants share a common layer: Funplay Skills, a set of 12 skills that standardize common workflows across engines. Skills provide reusable, engine-agnostic recipes — things like "create a 2D platformer controller" or "set up a UI health bar" — that the MCP tools execute in engine-specific ways.

This means switching from Cocos to Unity or Godot does not require relearning the high-level workflow. You change engines, keep the same skill invocations, and the MCP server translates.

Final Recommendation

Start with the core profile. Let your AI agent learn the 19 high-signal tools first. Expand to the full 67 when you hit a wall that execute_javascript cannot solve on its own.

Read more about Funplay's MCP ecosystem at gamebooom.ai/en/blog/, explore the Cocos extension at github.com/FunplayAI/funplay-cocos-mcp, and check sibling projects like FunplayAI/funplay-unity-mcp and FunplayAI/funplay-godot-mcp if your stack runs on Unity or Godot instead.