Use Gameye from AI agents (MCP server and skills)

Gameye has two open-source tools for AI agents. Both are MIT licensed and work with any agent that supports them.

Tool	What it does	Where
gameye-mcp	A Model Context Protocol server. Your agent can launch and stop sessions, check image tags and regions, read logs, download artifacts, and create applications and tokens through the Gameye API.	npm, `ghcr.io/gameye/gameye-mcp`, GitHub, MCP Registry as `io.github.Gameye/gameye-mcp`
gameye-skills	Twelve Agent Skills that teach a coding agent how to design an authoritative server, build it for Unity, Unreal, Godot or Node.js, containerize it, test it locally and run it on Gameye.	GitHub

The MCP server is in beta (0.x). Tool names, inputs and outputs may change between releases.

Before you start

A Gameye API token. Sign up at trial.gameye.com, then create a token in the Admin Panel under Users → API tokens.
Node.js 22 or newer for the MCP server, or Docker if you’d rather run the container image.

Claude Code: install the plugin

The plugin installs the twelve skills and the MCP server together, and keeps your token in Claude Code’s secure storage. In a Claude Code session:

/plugin marketplace add gameye/gameye-skills
/plugin install gameye@gameye

Claude Code asks for your API token and environment when you enable the plugin.

Any MCP client: add the server

With Claude Code and no plugin:

claude mcp add gameye --env GAMEYE_API_TOKEN=<your-token> -- npx -y gameye-mcp

Claude Desktop, Cursor, Windsurf and most other clients take an mcpServers block:

{
  "mcpServers": {
    "gameye": {
      "command": "npx",
      "args": ["-y", "gameye-mcp"],
      "env": {
        "GAMEYE_API_TOKEN": "<your-token>",
        "GAMEYE_ENV": "sandbox"
      }
    }
  }
}

Restart the client after you change its configuration. On Windows, many clients can’t start npx directly. Use "command": "cmd" with "args": ["/c", "npx", "-y", "gameye-mcp"]. The README covers Windows and Docker in full.

Environments

`GAMEYE_ENV`	API host	For
`sandbox` (default)	`api.sandbox-gameye.gameye.net`	Self-serve customers. Sandbox runs live matches, not only tests.
`production`	`api.production-gameye.gameye.net`	Enterprise customers.

The environment is fixed when the server starts. No tool can switch it, so an agent can’t move to production partway through a conversation.

Give the agent the smallest token it needs

Recipe	Scopes	The agent can
Observer	`session:read` `logs:read` `artifact:read` `regions:read` `region:read` `application:read`	Inspect sessions, logs, artifacts, regions and applications
Operator	Observer + `session:start` `session:stop` `session:write`	Also launch and stop sessions
Deployer	Operator + `application:write`	Also create and edit applications
Token admin	Deployer + `token:write`	Also create API tokens

Never pass your Gameye API token into a game server’s environment. The MCP server refuses a launch that would.

What the agent can do

Tool	What it does
`gameye_context`	Shows the environment, host, server version and guardrail defaults
`gameye_launch_session`	Starts a game server session (billable)
`gameye_list_sessions`, `gameye_describe_session`	List sessions, or get one session’s ports and players
`gameye_stop_session`	Stops one session by its exact ID
`gameye_player_join`, `gameye_player_leave`	Register players joining or leaving a session
`gameye_get_locations`, `gameye_list_regions`	Show where an image can run right now, and the regions your organization can use
`gameye_check_tag`, `gameye_list_registry_tags`, `gameye_enable_tag`	Check an image tag, list an application’s registry tags, and enable a tag so Gameye preloads it
`gameye_get_logs`, `gameye_get_artifacts`	Fetch session logs, or download files from a session container
`gameye_list_applications`, `gameye_create_application`, `gameye_edit_application`	Manage applications
`gameye_create_token`	Creates an API token and writes it to a file, never into the chat
`gameye_update_skills`	Checks the installed Gameye skills against the latest pack and updates them on request

The tools cover every operation in the Gameye client API.

Guardrails

Dry run. Launch, stop, enable tag and the application and token tools accept dryRun: true. The tool returns the exact request with the token masked and sends nothing.
Sessions end on their own. A launch without a ttl gets 30 minutes. Every agent launch carries the label launched-by: gameye-mcp, so you can find those sessions later.
One stop at a time. gameye_stop_session takes one exact session ID. There is no bulk stop.
The token stays out of results. Every tool result, errors included, shows the API token as ***.

Install only the skills

For Cursor, Copilot or any agent that reads SKILL.md files:

npx skills add gameye/gameye-skills --skill "*"

Add -g to install them for every project. The skills use the MCP tools when the server is connected, and plain HTTP calls to the API when it isn’t.

Example prompts

“Containerize my Godot server and get it running on Gameye in Frankfurt.”
“List my running sessions and show the logs of the one that crashed.”
“Which regions can run my image right now?”
“Create a read-only token for our dashboard and save it to ~/secrets/gameye-observer.”

Getting started
API V2 overview
Deploy from GitHub Actions
Machine-readable docs: llms.txt and llms-full.txt