Prerequisites

Before you start

The AI tooling is part of the standard INGenious distributable. There is nothing extra to install for the MCP server itself — but INGenie needs a model provider, and some tools need the usual INGenious runtimes.

1. Runtime

Requirement	Notes
`Java 17+`	Same requirement as the rest of INGenious. Verify with `java -version`.
`ingenious launcher`	`ingenious` / `ingenious.bat` / `ingenious.command` must be on your `PATH`, or referenced by an absolute path.
`A project`	The tools operate on an INGenious project folder. Create one with the IDE or `ingenious project create`.

2. Model provider

The MCP server itself needs no model — it only exposes tools. The model is supplied by whatever is driving it.

External MCP clientINGenie (IDE panel / ingenious ai)

The client brings its own model and subscription:

GitHub Copilot in VS Code
Claude Desktop
Cursor
Any client speaking JSON-RPC 2.0 over stdio

Configured through ~/.ingenious/ai.json:

{
  "provider": "copilot",
  "model": "gpt-4o",
  "baseUrl": "https://api.openai.com/v1",
  "apiKeyEnv": "OPENAI_API_KEY",
  "mode": "unattended"
}

provider	Authentication
`copilot` (default)	GitHub device flow. Run `/login` in the REPL, or sign in from the INGenie panel. A browser opens, you paste the code — no token to manage by hand.
`copilot-sdk`	Reuses the GitHub Copilot CLI login on your machine (`copilot auth login`). The INGenious MCP server is exposed to it automatically — see setup below.
`openai`	Any OpenAI-compatible chat-completions endpoint (OpenAI, Azure OpenAI, a local gateway). Set `baseUrl`, `model` and put the key in the environment variable named by `apiKeyEnv`.

Setting up the `copilot-sdk` provider

The copilot-sdk provider does not sign in by itself — it drives the official GitHub Copilot CLI installed on your machine and reuses the CLI's own authentication. Install and log in once, and every INGenious session picks it up.

1. Install the Copilot CLI2. Log in3. Point INGenious at it

The CLI ships as an npm package, so you need Node.js 22+ and npm 10+ first:

node --version
npm --version

Then install it globally:

npm install -g @github/copilot

If your .npmrc sets ignore-scripts=true, install with:

npm_config_ignore_scripts=false npm install -g @github/copilot

Verify that the copilot binary is on your PATH:

copilot --version

Authenticate the CLI from your normal shell:

copilot auth login

A browser window opens for the GitHub device flow — approve the request, and the token is stored by the CLI itself. You can also start copilot and run /login inside it.

Check the result at any time:

copilot auth status

/login inside INGenie

Running /login in the INGenious REPL while copilot-sdk is selected only reminds you to use the CLI — INGenious never stores a Copilot token for this provider.

In ~/.ingenious/ai.json:

{
  "provider": "copilot-sdk",
  "model": "claude-sonnet-4.5",
  "mode": "unattended"
}

Or switch at runtime from the REPL:

/model provider copilot-sdk
/model list

/model list queries the CLI for the models your subscription allows and lets you pick one. When the provider starts, INGenious automatically exposes its own MCP server to the CLI, so all ingenious_* tools are available to the model.

Requires an active Copilot subscription

The CLI needs Copilot Pro, Pro+, Business or Enterprise. If startup fails with "Could not start the Copilot CLI via the SDK", re-check copilot --version and copilot auth login in the same shell that launches INGenious.

Never commit keys

The openai provider deliberately reads its key from an environment variable rather than from ai.json. Keep it that way — ai.json lives in your home folder and is easy to leak.

export OPENAI_API_KEY="..."

3. Optional runtimes

Only needed by the corresponding tool families:

Runtime	Needed for
`Playwright browsers`	Browser discovery sessions and browser test execution. Downloaded on first use.
`Appium`	Mobile test execution.
`k6`	Performance tooling (`ingenious_perf_*`).
`Node.js`	The GitHub Copilot CLI (`copilot-sdk` provider, needs Node 22+), and building the VS Code extension from source.

One command to check them all

Ask INGenie to "run doctor" — or call the ingenious_doctor tool / ingenious doctor command. It reports the detected JDK, Playwright browsers, drivers and k6 in one go.

4. Proxy / network

The device-flow sign-in and any hosted model endpoint need outbound HTTPS. In a corporate network, make sure the usual Java proxy settings are honoured before reporting a sign-in failure.

Next up :

INGenie MCP Server