Conduit MCP server
Commit Oct 27, 2026Enables AI coding agents to interact with the Unity editor through a robust and context-efficient MCP server.
Also known as apkd/Conduit
- Stars
- 3
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 27, 2026
- License
- MIT
What is Conduit MCP?
Conduit enables AI coding agents to interact with the Unity editor through a robust and context-efficient MCP server.
Summary generated from the project's README
About Conduit MCP server
Conduit provides a way for AI coding agents to interact directly with the Unity editor. It is designed to be context-efficient, using a small number of versatile tools to conserve the agent's context window and save tokens. The server is built to be resilient, meaning it can survive assembly reloads, restarts, and crashes while supporting multiple agents and Unity instances simultaneously.
Developers can set up the server using a single Unity package and a server executable. An automated setup wizard is available within the Unity editor under Preferences to help configure code editors. The system works on both Windows and Linux, supporting both the editor and development builds. It can be connected via stdio or through an HTTP server, and it is compatible with clients like Claude Code, Claude Desktop, and Codex.
What you can do with it
- 1Automate Unity editor tasksAn AI agent can perform various editor operations such as saving scenes, refreshing the asset database, or reimporting assets.
- 2Analyze project structure and dependenciesA developer can ask an AI to find missing scripts or identify all references to a specific object within the project.
- 3Monitor performance and run testsYou can instruct an AI client to browse the profiler or execute tests in edit mode or play mode.
- 4Execute code within the editorAn AI agent can run specific code snippets to manipulate the Unity environment directly.
Written from the project's README. Every use case is checked against a quote from it.
Configuration
CONDUIT_IPC_ROOTXDG_RUNTIME_DIR
| Name | Description | |
|---|---|---|
| CONDUIT_IPC_ROOT | — | |
| XDG_RUNTIME_DIR | — |
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableapkd/conduit is public and reachablegithub.com/apkd/conduit ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-27github.com/apkd/conduit ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub ▄████████ ▄██████▄ ███▄▄▄▄ ████████▄ ███ █▄ ▄█ ███
███ ███ ███ ███ ███▀▀▀██▄ ███ ▀███ ███ ███ ███ ▀█████████▄
███ █▀ ███ ███ ███ ███ ███ ███ ███ ███ ███▌ ▀███▀▀██
███ ███ ███ ███ ███ ███ ███ ███ ███ ███▌ ███ ▀
███ ███ ███ ███ ███ ███ ███ ███ ███ ███▌ ███
███ █▄ ███ ███ ███ ███ ███ ███ ███ ███ ███ ███
███ ███ ███ ███ ███ ███ ███ ▄███ ███ ███ ███ ███
████████▀ ▀██████▀ ▀█ █▀ ████████▀ ████████▀ █▀ ▄████▀
A simple and fast MCP server for Unity.
A Unity MCP server that stays out of the way of your coding agent.
- Robust: survives crashes, restarts, assembly reloads, and handles multiple agents and Unity instances.
- Context-efficient: conserves the agent's context window and saves tokens. Small number of versatile tools.
- Simple setup: one Unity package, one server exe, automated setup wizard. No dependencies, no pollution.
- Supports Linux and Windows, both in the editor and in development builds.
[!WARNING] Granting an AI agent access to Unity indirectly gives them escalated access to your machine. Agents may be able to perform actions outside the regular sandbox through Unity. Have a backup strategy, and make sure your work machine is resilient to data loss.
Installation
1. Install the Unity package
- Window ⟶ Package Manager ⟶
+⟶ Install package from git URL - Paste this URL (the
releasebranch points at the latest commit that passed all tests)
https://github.com/apkd/Conduit.git?path=/Conduit.Unity#release
- Or you can also declare it directly in
Packages/manifest.jsoninstead:
"dependencies": {
"dev.tryfinally.conduit": "https://github.com/apkd/Conduit.git?path=/Conduit.Unity#release",
2a. Just use the setup wizard
[!TIP] The Unity package includes a wizard for setting up the MCP server. This lets you download the server executable and configure your installed code editors in just a few clicks.
You can find the setup wizard in Unity here:
Edit → Preferences... → Conduit
2b. Install the MCP server
[!TIP] If you used the setup wizard, you can skip this step.
First, you need to grab the MCP server executable, and put it in your project (for a project-local installation) or some stable location on your computer (for a user-level setup).
Download conduit-win-x64.exe and move it to the location where you want to keep it.
For example, this PowerShell command installs it as conduit.exe in your user profile:
New-Item -ItemType Directory -Force "$HOME\Conduit"
Invoke-WebRequest "https://github.com/apkd/Conduit/releases/latest/download/conduit-win-x64.exe" -OutFile "$HOME\Conduit\conduit.exe"
Unblock-File "$HOME\Conduit\conduit.exe"
Download conduit-linux-x64, place it in the location where you want to keep it, and make it executable.
For example, these commands install it as conduit in ~/.local/bin:
mkdir -p "$HOME/.local/bin"
curl -fL "https://github.com/apkd/Conduit/releases/latest/download/conduit-linux-x64" -o "$HOME/.local/bin/conduit"
chmod +x "$HOME/.local/bin/conduit"
Conduit supports Linux x64 and includes .NET. With nix-command and flakes
enabled, you can try it or install it:
nix run github:apkd/Conduit -- --version
nix profile add github:apkd/Conduit#conduit
For your NixOS or Home Manager configuration, add Conduit to your flake:
inputs.conduit.url = "github:apkd/Conduit";
Put inputs.conduit.packages.x86_64-linux.default in environment.systemPackages
or home.packages, then rebuild. Configure your MCP client to run conduit.
If you'd like an HTTP server, add these entries to your Home Manager module list:
inputs.conduit.homeManagerModules.default
{
services.conduit = {
enable = true;
};
}
Apply your Home Manager configuration, then point your MCP client at
http://127.0.0.1:5080. The service starts and stops automatically with your desktop
session. You can change the port with services.conduit.port. Authentication is
optional. To enable it, set services.conduit.tokenFile to a token file path as
described below and configure the client to send it as a bearer token.
Alternatively, set tokenSha256File to the absolute path of a token fingerprint
file as described below. Set protectProcess = true on Linux to protect Conduit's
memory and turn off its .NET diagnostics.
Install Unity Hub (unityhub) separately if you'd like Conduit to launch Unity.
You can connect to an editor you've already opened without Hub.
To update a flake installation, run nix flake update conduit and rebuild your
configuration. For profile installations, run nix profile upgrade <entry>;
nix profile list shows the entry name.
Install the .NET 10 SDK for your operating system. Then clone this repository and publish the server:
dotnet publish Conduit.Server/Conduit.csproj -c Release
The executable is written to Conduit.Server/publish/win-x64/conduit.exe on Windows or Conduit.Server/publish/linux-x64/conduit on Linux.
3. Configure your code editor
[!TIP] If you used the setup wizard, you can skip this step.
Replace the executable paths in the examples below with the path where you installed the MCP server.
Configure the MCP server in either location:
- Unity project:
.codex/config.toml. Codex loads this file after the project is trusted. - User account:
%USERPROFILE%\.codex\config.tomlon Windows, or~/.codex/config.tomlon Linux.
stdio | Windows (Native)
[mcp_servers.unity]
command = "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe"
args = []
cwd = "C:\\src\\Conduit"
disabled_tools = []
tool_timeout_sec = 300
enabled = true
stdio | Windows (WSL)
[mcp_servers.unity]
command = "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe"
args = []
cwd = "/mnt/c/src/Conduit"
disabled_tools = []
tool_timeout_sec = 300
enabled = true
stdio | Linux
[mcp_servers.unity]
command = "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit"
args = []
env_vars = ["XDG_RUNTIME_DIR"]
cwd = "/home/you/src/Conduit"
disabled_tools = []
tool_timeout_sec = 300
enabled = true
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
codex mcp add unity --url http://127.0.0.1:5080
approve tool calls
To avoid going insane from having to approve every tool call separately:
[mcp_servers.unity]
tools.discard_scenes.approval_mode = "approve"
tools.execute_code.approval_mode = "approve"
tools.find_missing_scripts.approval_mode = "approve"
tools.find_references_to.approval_mode = "approve"
tools.from_json_overwrite.approval_mode = "approve"
tools.get_dependencies.approval_mode = "approve"
tools.help.approval_mode = "approve"
tools.playmode.approval_mode = "approve"
tools.editmode.approval_mode = "approve"
tools.profiler_browse.approval_mode = "approve"
tools.profiler_overview.approval_mode = "approve"
tools.profiler_record.approval_mode = "approve"
tools.project_settings.approval_mode = "approve"
tools.record.approval_mode = "approve"
tools.refresh_asset_database.approval_mode = "approve"
tools.reimport_assets.approval_mode = "approve"
tools.reflect.approval_mode = "approve"
tools.restart.approval_mode = "approve"
tools.run_tests_editmode.approval_mode = "approve"
tools.run_tests_player.approval_mode = "approve"
tools.run_tests_playmode.approval_mode = "approve"
tools.save_scenes.approval_mode = "approve"
tools.screenshot.approval_mode = "approve"
tools.search.approval_mode = "approve"
tools.show.approval_mode = "approve"
tools.status.approval_mode = "approve"
tools.to_json.approval_mode = "approve"
tools.view_burst_asm.approval_mode = "approve"
tools.view_il.approval_mode = "approve"
Configure the MCP server in either location:
- Unity project:
.mcp.json. - User account:
%USERPROFILE%\.claude.jsonon Windows, or~/.claude.jsonon Linux.
The commands below create the Unity project configuration. Replace --scope project with --scope user to configure the User account instead.
stdio | Windows (Native)
claude mcp add --scope project --transport stdio unity -- C:\src\Conduit\Conduit.Server\publish\win-x64\conduit.exe
stdio | Windows (WSL)
claude mcp add --scope project --transport stdio unity -- /mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe
stdio | Linux
claude mcp add --scope project --transport stdio unity -- /home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
Add the server to your private user configuration:
{
"mcpServers": {
"unity": {
"type": "http",
"url": "http://127.0.0.1:5080"
}
}
}
Claude Desktop keeps one MCP configuration for the User account:
- Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json, or the equivalent path underXDG_CONFIG_HOME
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": []
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"args": []
}
}
}
http
Claude Desktop's Custom Connectors are reached through Anthropic's cloud and cannot connect directly to a server on localhost.
Use the local stdio configuration above for Conduit.
Cursor uses mcp.json with a top-level mcpServers object.
Save it in either location:
- Unity project:
.cursor/mcp.json. - User account:
%USERPROFILE%\.cursor\mcp.jsonon Windows, or~/.cursor/mcp.jsonon Linux.
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": []
}
}
}
stdio | Windows (WSL)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe",
"args": []
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"args": []
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcpServers": {
"unity": {
"url": "http://127.0.0.1:5080"
}
}
}
Windsurf keeps one MCP configuration for the User account:
- Windows:
%USERPROFILE%\.codeium\mcp_config.json - Linux:
~/.codeium/mcp_config.json
Some older Cascade builds use %USERPROFILE%\.codeium\windsurf\mcp_config.json on Windows or ~/.codeium/windsurf/mcp_config.json on Linux.
Use Windsurf Settings → Cascade → MCP Servers → View Raw Config when both files exist.
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": []
}
}
}
stdio | Windows (WSL)
{
"mcpServers": {
"unity": {
"command": "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe",
"args": []
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"args": []
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcpServers": {
"unity": {
"serverUrl": "http://127.0.0.1:5080"
}
}
}
Cline keeps one MCP configuration for the User account. The location depends on which version of Cline you use.
For the VS Code extension:
- Windows:
%APPDATA%\Code\User\globalStorage\saoudrizwan.claude-dev\settings\cline_mcp_settings.json - Linux:
~/.config/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json, or the equivalent path underXDG_CONFIG_HOME
For JetBrains and CLI builds:
- Windows:
%USERPROFILE%\.cline\data\settings\cline_mcp_settings.json - Linux:
~/.cline/data/settings/cline_mcp_settings.json
CLINE_MCP_SETTINGS_PATH, CLINE_DATA_DIR, and CLINE_DIR can relocate the JetBrains and CLI file.
Current Cline uses flat transport fields in each server entry.
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": [],
"disabled": false
}
}
}
stdio | Windows (WSL)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe",
"args": [],
"disabled": false
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"args": [],
"env": { "XDG_RUNTIME_DIR": "${env:XDG_RUNTIME_DIR}" },
"disabled": false
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcpServers": {
"unity": {
"type": "streamableHttp",
"url": "http://127.0.0.1:5080",
"disabled": false
}
}
}
Kilo Code's extension and CLI share the unified kilo.json/kilo.jsonc format.
Configure the MCP server in either location:
- Unity project:
kilo.json,kilo.jsonc,.kilo/kilo.json, or.kilo/kilo.jsonc. - User account:
%USERPROFILE%\.config\kilo\kilo.jsonon Windows, or~/.config/kilo/kilo.jsonon Linux. The.jsoncfilename also works.
On Linux, XDG_CONFIG_HOME can move the User account file.
stdio | Windows (Native)
{
"mcp": {
"unity": {
"type": "local",
"command": ["C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe"],
"enabled": true
}
}
}
stdio | Windows (WSL)
{
"mcp": {
"unity": {
"type": "local",
"command": ["/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe"],
"enabled": true
}
}
}
stdio | Linux
{
"mcp": {
"unity": {
"type": "local",
"command": ["/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit"],
"enabled": true
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcp": {
"unity": {
"type": "remote",
"url": "http://127.0.0.1:5080",
"enabled": true
}
}
}
Continue's IDE extensions load MCP files from either location:
- Unity project:
.continue/mcpServers/. - User account:
%USERPROFILE%\.continue\mcpServers\on Windows, or~/.continue/mcpServers/on Linux.
Continue accepts native schema-v1 YAML and Claude-compatible JSON files in either folder.
The cn CLI does not currently auto-discover files from either directory.
Create unity.yaml in either directory:
stdio | Windows (Native)
name: Unity MCP
version: 0.0.1
schema: v1
mcpServers:
- name: unity
type: stdio
command: C:\src\Conduit\Conduit.Server\publish\win-x64\conduit.exe
cwd: C:\src\Conduit
stdio | Windows (WSL)
name: Unity MCP
version: 0.0.1
schema: v1
mcpServers:
- name: unity
type: stdio
command: /mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe
cwd: /mnt/c/src/Conduit
stdio | Linux
name: Unity MCP
version: 0.0.1
schema: v1
mcpServers:
- name: unity
type: stdio
command: /home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit
cwd: /home/you/src/Conduit
env:
XDG_RUNTIME_DIR: ${{ secrets.XDG_RUNTIME_DIR }}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
name: Unity MCP
version: 0.0.1
schema: v1
mcpServers:
- name: unity
type: streamable-http
url: http://127.0.0.1:5080
Configure the MCP server in either location:
- Unity project:
opencode.json,opencode.jsonc,.opencode/opencode.json, or.opencode/opencode.jsonc. - User account:
%USERPROFILE%\.config\opencode\opencode.jsonon Windows, or~/.config/opencode/opencode.jsonon Linux. The.jsoncfilename also works.
On Linux, XDG_CONFIG_HOME can move the User account file.
Local MCP commands are arrays whose first element is the executable path.
stdio | Windows (Native)
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"unity": {
"type": "local",
"command": ["C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe"],
"enabled": true
}
}
}
stdio | Windows (WSL)
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"unity": {
"type": "local",
"command": ["/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe"],
"enabled": true
}
}
}
stdio | Linux
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"unity": {
"type": "local",
"command": ["/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit"],
"enabled": true
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"unity": {
"type": "remote",
"url": "http://127.0.0.1:5080",
"enabled": true
}
}
}
Configure the MCP server in either location:
- Unity project:
.gemini/settings.json. - User account:
%USERPROFILE%\.gemini\settings.jsonon Windows, or~/.gemini/settings.jsonon Linux.
Setting GEMINI_CLI_HOME moves the User account file to a .gemini folder inside that directory.
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"cwd": "C:\\src\\Conduit"
}
}
}
stdio | Windows (WSL)
{
"mcpServers": {
"unity": {
"command": "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe",
"cwd": "/mnt/c/src/Conduit"
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"cwd": "/home/you/src/Conduit"
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcpServers": {
"unity": {
"type": "http",
"url": "http://127.0.0.1:5080"
}
}
}
Configure the MCP server in either location:
- Unity project:
.github/mcp.jsonor.mcp.json. - User account:
%USERPROFILE%\.copilot\mcp-config.jsonon Windows, or~/.copilot/mcp-config.jsonon Linux.
Setting COPILOT_HOME moves the User account file to that directory.
stdio | Windows (Native)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": [],
"env": {},
"tools": ["*"]
}
}
}
stdio | Windows (WSL)
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/mnt/c/src/Conduit/Conduit.Server/publish/win-x64/conduit.exe",
"args": [],
"env": {},
"tools": ["*"]
}
}
}
stdio | Linux
{
"mcpServers": {
"unity": {
"type": "stdio",
"command": "/home/you/src/Conduit/Conduit.Server/publish/linux-x64/conduit",
"args": [],
"env": { "XDG_RUNTIME_DIR": "${XDG_RUNTIME_DIR}" },
"tools": ["*"]
}
}
}
http
conduit --http [--port 5080] [--url http://127.0.0.1:5080]
{
"mcpServers": {
"unity": {
"type": "http",
"url": "http://127.0.0.1:5080",
"tools": ["*"]
}
}
}
Interactive alternative:
/mcp add
Visual Studio uses mcp.json with a top-level servers object.
Configure the MCP server in either location:
- Unity project:
.mcp.json. - User account:
%USERPROFILE%\.mcp.json.
Visual Studio also recognizes .vs/mcp.json, .vscode/mcp.json, and .cursor/mcp.json inside the Unity project folder.
stdio | Windows (Native)
{
"servers": {
"unity": {
"type": "stdio",
"command": "C:\\src\\Conduit\\Conduit.Server\\publish\\win-x64\\conduit.exe",
"args": []
}
}
}
Conduit MCP questions
What operating systems does Conduit support?
Conduit supports both Linux and Windows, working in both the Unity editor and development builds.
How do I set up Conduit in Unity?
You can use the built-in setup wizard located in Unity under Edit → Preferences... → Conduit to download the executable and configure your editors.
Is there a risk to using Conduit with an AI agent?
Yes, granting an AI agent access to Unity can indirectly give them escalated access to your machine, potentially allowing actions outside the regular sandbox.