---
name: install-sparklaunch
description: Install or connect SparkLaunch from its official skills repository using the package for the user's AI client. Use when the user asks to set up SparkLaunch in ChatGPT, Codex, Claude Code, Cursor, Gemini CLI, or Muse Code.
---

# Install SparkLaunch

Set up SparkLaunch for the user's AI client using the official repository:

**https://github.com/SparkLaunch-Dev/SparkLaunch-Skills**

This is a standalone installer skill. It contains everything needed to choose an
installation route; it does not bundle the SparkLaunch plugins. A user can give
this file to their assistant and say: **"Install SparkLaunch for the platform I
am using."**

## Choose the client and current package

1. Identify the AI client from trusted runtime information or the user's stated
   preference. Windows, macOS, and Linux are operating systems, not plugin
   targets. Distinguish ChatGPT from Codex, Claude Code from Claude web, and
   Gemini CLI from Gemini web or Antigravity. Ask which client only if unclear.
2. Read the repository's current [README](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/README.md),
   [release state](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/release-state.json),
   and the guide for the selected client below. Check the installed client's
   capabilities before using version-dependent commands. Respect a requested
   version or Git ref; otherwise use the repository's documented current route.
3. Inspect any existing SparkLaunch installation. Reuse it if it already meets
   the request. For an update, refresh only the SparkLaunch source/package and
   verify its version. Do not uninstall unrelated plugins or replace existing
   configuration files.
4. Perform the requested installation with available native tools or shell
   access. If the environment cannot install software, provide the exact steps
   for the selected client and identify the step the user must complete.

| AI client | Package in the repository | Setup guide |
| --- | --- | --- |
| ChatGPT / Codex | `plugins/sparklaunch/` | [OpenAI connection](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/adapters/openai/connection.md) |
| Claude Code | `plugins/claude/sparklaunch/` | [Claude installation](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/adapters/claude/DISTRIBUTION.md) |
| Cursor | `plugins/cursor/sparklaunch/` | [Cursor installation](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/adapters/cursor/DISTRIBUTION.md) |
| Gemini CLI | `plugins/gemini/sparklaunch/` | [Gemini installation and eligibility](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/adapters/gemini/README.md) |
| Muse Code | `plugins/muse/sparklaunch/skills/` | [Muse skills installation](https://github.com/SparkLaunch-Dev/SparkLaunch-Skills/blob/main/adapters/muse/README.md) |

Clone the repository only when the selected route needs local files. Use a new
directory or an existing verified checkout without overwriting local work:

```sh
git clone https://github.com/SparkLaunch-Dev/SparkLaunch-Skills.git
```

The paths in the commands below assume the shell is in that checkout. Adapt
paths and quoting to the user's operating system. Install the complete generated
host package; copying this installer file alone does not install its tools.

## Codex

Register the repository as a Git-backed marketplace, then install **sparklaunch**
from **sparklaunch-skills**. The catalog is `.agents/plugins/marketplace.json`
and points to `plugins/sparklaunch/`.

If the installed CLI supports `codex plugin marketplace add`, run:

```sh
codex plugin marketplace add SparkLaunch-Dev/SparkLaunch-Skills --ref main
```

Then open `/plugins` in Codex CLI and install the SparkLaunch entry. On a desktop
client with **Add plugin marketplace**, use the repository URL and install the
same entry. If sparse checkout options are offered, include both `.agents/plugins`
and `plugins` so the package is available. Substitute the user's requested ref
when applicable. Start a fresh task after installation.

If plugin support is absent, explain the required supported client/update;
do not invent a CLI install command or silently substitute a skills-only install.
See [OpenAI plugin packaging](https://developers.openai.com/plugins/build/plugins)
and [plugin installation](https://learn.chatgpt.com/docs/plugins).

## ChatGPT

Open **Plugins**, find the official SparkLaunch entry if available to this user,
install it, and complete the connection prompt. Start a new conversation with
SparkLaunch selected. Confirm availability in the actual account rather than
assuming a public listing from the GitHub repository.

If SparkLaunch is unavailable, a workspace administrator can use **Admin >
Plugins > Add > Import marketplace** with the repository URL as Source, an empty
Path, and the intended branch or ref. Workspace import is a separate administrator
task; perform it only when the user requests that scope. Imported packages that
declare MCP servers are **Desktop only**, even with a remote HTTPS endpoint.
Explain that boundary to web/mobile users and name the access requirement if
they cannot install the package.

See [OpenAI plugin installation](https://learn.chatgpt.com/docs/plugins) and
[workspace marketplace import](https://learn.chatgpt.com/docs/enterprise/plugin-management).

## Claude Code

In an interactive Claude Code session, run these in order:

```text
/plugin marketplace add SparkLaunch-Dev/SparkLaunch-Skills
/plugin install sparklaunch@sparklaunch-skills
```

These are Claude Code slash commands, not shell commands. Choose the user's
requested scope, or user scope for a personal installation. Run `/mcp`, complete
SparkLaunch browser sign-in if prompted, then reload plugins or start a fresh
session. Claude web connector setup is a separate route; do not claim these
commands install a Claude web plugin.

## Cursor

Use the official SparkLaunch Marketplace entry if it is available. Otherwise
install the repository package locally: copy the **contents** of
`plugins/cursor/sparklaunch/` to `~/.cursor/plugins/local/sparklaunch/`, including
`plugin.json`, `mcp.json`, and `skills/`. Resolve `~` to the user's home directory
on their OS. Preserve any existing installation and configuration.

Run **Developer: Reload Window** or restart Cursor. Open **Customize** and verify
the plugin and MCP server, then follow the authentication prompt. Local imports
may require administrator enablement; a same-named Marketplace installation takes
precedence. Use a copy: Cursor skips symlinks pointing outside its local plugins
directory. See [Cursor's local plugin instructions](https://cursor.com/docs/plugins#test-plugins-locally).

## Gemini CLI

Check the adapter guide's current account eligibility first. This package targets
Gemini CLI; it does not establish support for Gemini web or Antigravity.

Use the repository URL installer only after verifying a published release has a
compatible Gemini archive with `gemini-extension.json` at its root:

```sh
gemini extensions install https://github.com/SparkLaunch-Dev/SparkLaunch-Skills
```

If no compatible release is available, install the generated package from a
local checkout instead:

```sh
gemini extensions install ./plugins/gemini/sparklaunch
gemini extensions list
```

Restart Gemini CLI, then run `/mcp` and `/mcp auth sparklaunch` in its interactive
session. Extension installation commands run in the shell. Verify discovery after
restarting. See [Gemini extension installation](https://geminicli.com/docs/extensions/reference/).

## Muse Code

Install the generated skills individually. For each skill directory under
`plugins/muse/sparklaunch/skills/`, validate it, then install it at user scope:

```text
muse skills validate ./plugins/muse/sparklaunch/skills/sparklaunch-platform
muse skills install ./plugins/muse/sparklaunch/skills/sparklaunch-platform --scope user
```

Repeat for every skill directory present in that package, then run
`muse skills list`. Use `muse skills inspect sparklaunch-platform` to verify an
installed skill. Follow an explicitly requested project scope instead when
supported by the current guide.

**This adapter provides skills only; protected SparkLaunch MCP access stays
disabled.** Do not enable its example server or add static credentials as a
workaround. If the user needs connected tools, explain the limitation and let
them choose a host with supported OAuth.

## Connect and verify

Enabled packages use Streamable HTTP at **https://sparklaun.ch/api/mcp/**.
Let the host manage SparkLaunch sign-in and permission consent. Never request
passwords, tokens, API keys, OAuth codes, client secrets, or authorization headers
in chat or package files.

After the host reports connection success, verify that the SparkLaunch skills
and tools are loaded in a fresh session. Use the read-only `projects.list` tool,
or its host-exposed equivalent, to check access. If there are no accessible
projects, report that result without creating one. Do not publish, send messages,
change project data, or purchase anything as an installation test.

Report the client, installed package/version or source ref, whether a restart is
needed, and the observed connection result. Distinguish **installed**, **loaded**,
and **authenticated**; report any incomplete step plainly. For Muse, report
**skills installed; protected MCP unavailable**. Stop after successful verification.

For an unlisted client, check the current repository for an official adapter.
If none exists, explain that support is unconfirmed and provide the repository
link instead of installing another client's package. Support: **support@sparklaun.ch**.