tuanjie-mcp

tuanjie-mcp

Enables AI tools to connect to and control Unity/Tuanjie editors via the Codely Bridge TCP server, supporting project management, scene manipulation, and console reading.

Category
Visit Server

README

tuanjie-mcp

English | 中文

tuanjie-mcp is a stdio MCP server for the existing cn.tuanjie.codely.bridge TCP server. It runs outside Unity/Tuanjie and does not add or modify Unity package code.

The adapter supports multiple open editors. You can pin it to a process at startup, select an instance at runtime, or point it at a project/port. It does not depend on the MCP client's working directory.

Quick setup for AI tools

Build the server once, then register the same stdio command in your AI client's MCP settings:

cd /path/to/tuanjie-mcp
npm install
npm run build

The command is always node /absolute/path/to/tuanjie-mcp/dist/index.js. Use an absolute path so the client can start the server from any workspace.

Codex

Add this to ~/.codex/config.toml (the active local configuration uses this layout):

[mcp_servers.tuanjie-mcp]
command = "node"
args = ["/absolute/path/to/tuanjie-mcp/dist/index.js"]

Or register the identical server from the command line:

codex mcp add tuanjie-mcp -- node /absolute/path/to/tuanjie-mcp/dist/index.js

Cursor

Add this entry to ~/.cursor/mcp.json under mcpServers (the local Cursor configuration uses this layout):

{
  "mcpServers": {
    "tuanjie": {
      "command": "node",
      "args": ["/absolute/path/to/tuanjie-mcp/dist/index.js"],
      "cwd": "/absolute/path/to/tuanjie-mcp",
      "trust": true
    }
  }
}

Other common MCP clients

Any client that supports local stdio MCP servers can use this adapter, including Claude Desktop, VS Code MCP-capable extensions, Windsurf, Cline, Roo Code, OpenCode, and similar tools. In that client's MCP settings, create a local server and supply the same values:

Field Value
transport stdio
command node
arguments [/absolute/path/to/tuanjie-mcp/dist/index.js]
working directory optional; /absolute/path/to/tuanjie-mcp is safe

Client configuration file names and JSON/TOML schema vary, but the launch command does not. Restart or reload the client after saving its MCP settings.

Requirements

  • Node.js 20 or newer
  • A Unity/Tuanjie project with cn.tuanjie.codely.bridge running
  • lsof on macOS/Linux when selecting by PID (Get-NetTCPConnection is used on Windows)

Install and build

cd /path/to/tuanjie-mcp
npm install
npm run build

Select one editor process

Pin the server to a running Unity/Tuanjie PID:

node /path/to/tuanjie-mcp/dist/index.js --unity-pid 57250

Equivalent MCP client configuration:

{
  "mcpServers": {
    "tuanjie": {
      "command": "node",
      "args": [
        "/path/to/tuanjie-mcp/dist/index.js",
        "--unity-pid",
        "57250"
      ]
    }
  }
}

The PID resolver first maps the process's listening ports and verifies the Codely WELCOME UNITY-TCP ... FRAMING=1 handshake. unity_ping then calls manage_editor/get_project_root so a live but incorrect endpoint is not accepted as the intended project.

You can also configure a project without setting the MCP process cwd:

{
  "mcpServers": {
    "tuanjie": {
      "command": "node",
      "args": [
        "/path/to/tuanjie-mcp/dist/index.js"
      ],
      "env": {
        "TUANJIE_UNITY_PROJECT": "/absolute/path/to/MyUnityProject"
      }
    }
  }
}

Startup selectors, in precedence order when supplied explicitly:

CLI Environment Meaning
--unity-pid 57250 TUANJIE_UNITY_PID=57250 Resolve the Bridge owned by one editor process
--unity-project /path/project TUANJIE_UNITY_PROJECT=/path/project Read the project's .com-unity-codely.json
--unity-port 50206 TUANJIE_UNITY_PORT=50206 Connect directly to a Bridge port
--unity-host 127.0.0.1 TUANJIE_UNITY_HOST=127.0.0.1 Override the Bridge host

When no selector is set, the server checks the current directory for backward compatibility, then discovers per-project status files under ~/.unity-tcp. One reachable instance is selected automatically. If several are reachable, the tool returns an explicit ambiguity error instead of guessing. Port 25916 remains the final compatibility fallback.

Runtime instance selection

The MCP client can switch editors without restarting the server:

  1. Call unity_list_instances to see project paths, ports, reachability, and PIDs when available.
  2. Call unity_select_instance with exactly one of pid, project, or port.
  3. Call unity_ping to see the selected endpoint and the project identity returned by the Bridge.

The selection persists for the lifetime of that MCP server process.

Switching instances through conversation

The screenshot below shows a typical workflow: with multiple Unity/Tuanjie editors open, the user simply asks the model to test the connection. The model discovers all running instances, selects the one matching the current repository, and pings it to confirm — all through natural-language conversation.

Switch project via LLM

You can ask the model to switch processes for you in plain language instead of calling the tools manually. The model performs the same two steps: it first calls unity_list_instances (which reports each instance's pid, projectRoot, port, and reachability), then calls unity_select_instance.

For example, in the chat you can say:

List all current Tuanjie processes, then switch to my Unity project.

or:

Switch to the Tuanjie process with PID 57250, then read the Console.

The model calls unity_list_instances, identifies the target from the returned projectRoot / pid, and then calls unity_select_instance. After a successful switch, subsequent Unity tool calls (unity_ping, read_console, manage_scene, etc.) are routed to that instance until you call unity_select_instance again or restart the MCP server.

Example unity_select_instance arguments:

{ "pid": 57250 }

Or by project path / port:

{ "project": "/path/to/MyUnityProject" }
{ "port": 50206 }

If your description is not enough to uniquely identify an instance — for example when two project names are similar — the model should list the candidate PIDs and project paths and ask you to confirm before switching.

Tools

  • unity_initialize_bridge — install the Bridge into a project that does not have it yet
  • unity_list_instances
  • unity_select_instance
  • unity_ping
  • unity_bridge_call — raw { type, params } passthrough
  • read_console
  • manage_editor
  • manage_scene
  • manage_gameobject
  • execute_menu_item
  • execute_csharp_script

The thin wrappers only map MCP arguments to the existing Bridge command. Bridge responses are returned unchanged, including { success, message, data, state, ... }. A Bridge error is marked as an MCP tool error while retaining the original JSON payload.

Initialize a project without Bridge

Call unity_initialize_bridge with a Unity/Tuanjie project root. If the project already lists cn.tuanjie.codely.bridge in Packages/manifest.json, the tool is a no-op. Otherwise it starts that project in batch mode and executes PackageManager.Client.Add("cn.tuanjie.codely.bridge"); it then verifies the resulting manifest and removes its temporary Editor bootstrap script.

On macOS, the tool automatically matches ProjectSettings/ProjectVersion.txt to /Applications/Tuanjie/Hub/Editor/<version>/Tuanjie.app/Contents/MacOS/Tuanjie. For another installation layout or platform, pass unity_executable or set TUANJIE_UNITY_EXECUTABLE.

Connection settings

  • TUANJIE_UNITY_TIMEOUT_MS / --timeout-ms: connect and frame timeout, default 10000
  • TUANJIE_UNITY_STATUS_DIR / --status-dir: override instance status discovery
  • UNITY_TCP_STATUS_DIR: legacy Bridge-compatible status directory override

The TCP client validates the welcome line and uses an unsigned 64-bit big-endian payload length followed by UTF-8 JSON. Connections are reused and calls are serialized. A failed connection is discarded so the next tool call can reconnect; commands are not automatically replayed because a lost response does not prove that a Unity-side write failed.

Verify

npm test
node dist/index.js --help

For a live editor, start the MCP server with a PID/project selector and call unity_ping, followed by read_console with action: "get" and a small count.

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured