Connect your tools

maximal connects the AI coding tools you already use to the models in your GitHub Copilot plan. Your tool works exactly as before — it’s just running on Copilot now. No extra API keys, no second bill.

The simplest way to connect a tool is a single switch. It takes about a minute.

Before you start

Two quick things first:

• Install maximal and leave it open so it keeps running in the background.
• Sign in with GitHub so maximal can use your Copilot plan.

Turn on a tool

1. Open maximal and go to the Apps section.
2. Find your tool and flip its switch on. maximal sets it up for you.
3. Start the tool — it’s now running on your Copilot models.

This works today for Claude Code and Claude Desktop. Flip the switch, open the tool, and use it the way you always have.

Claude Desktop (Cowork mode): after you switch it on, use Claude Desktop’s model picker as usual — maximal handles the model names behind the scenes. If you manage Claude Desktop through MDM, see the admin notes in docs/admin/claude-desktop-mdm.md.

Copilot CLI is listed in the Apps section too, but connecting it is coming soon — there’s no switch yet.

If a tool is already pointed somewhere else, maximal won’t overwrite it. It backs off and tells you about the conflict so nothing gets clobbered.

Next steps

• Watch your usage — Copilot quotas and token counts, live.
• Troubleshooting — if a tool won’t connect or Copilot rejects a request.

For developers

To connect a tool that isn’t in the Apps list — Codex, opencode, or your own SDK and HTTP clients — point it at maximal by hand. Once you’re signed in, maximal is listening locally at these addresses:

• Anthropic-compatible: http://localhost:4141
• OpenAI-compatible: http://localhost:4141/v1

The exact address and a ready-to-copy API key always live in the Endpoint section of the app. See Endpoint and API keys for the copy helpers.

Claude Code (manual)

If you’d rather set Claude Code yourself instead of using its switch, add these to your environment:

export ANTHROPIC_BASE_URL=http://localhost:4141
export ANTHROPIC_AUTH_TOKEN=your-maximal-key   # any value works, or mint a stable key
export ANTHROPIC_MODEL=<a model from the Models section>

The Models section lists everything available through your Copilot plan.

Codex

Codex isn’t auto-configured, so point it at maximal’s OpenAI-compatible address. In your Codex settings, use:

• Base URL: http://localhost:4141/v1
• API key: any maximal-minted key (see below)

Codex will behave as if it’s talking to api.openai.com. Grab the exact values from the Endpoint section.

opencode

Point opencode at the same OpenAI-compatible endpoint:

• Base URL: http://localhost:4141/v1
• API key: a maximal-minted key

Copy both straight from the Endpoint section and paste them into opencode’s provider settings.

Any Anthropic or OpenAI SDK

Any client that speaks the Anthropic or OpenAI wire format works. Set the base URL and send a maximal key as the auth header:

• Anthropic-style: base URL http://localhost:4141, key as x-api-key
• OpenAI-style: base URL http://localhost:4141/v1, key as Authorization: Bearer <key>

A quick sanity check with curl:

curl http://localhost:4141/v1/chat/completions \
  -H "Authorization: Bearer your-maximal-key" \
  -H "Content-Type: application/json" \
  -d '{"model":"<a model from Models>","messages":[{"role":"user","content":"hello"}]}'

The Endpoint section has copy-curl and copy-env-var helpers so you don’t have to type any of this.

A note on API keys

• Mint a stable key for anything long-running. In API clients, create a named key and use it in your tools. The auto-generated endpoint key rotates every time maximal restarts. So it’s fine for a quick test, but not for a tool you leave running.
• No keys configured? Then maximal accepts every local request without checking auth. That’s convenient, but worth knowing — mint a key if you want to lock local access down.

See Endpoint and API keys for the full details.