# Install Zigguratoken with your existing agent

**Zigguratoken is a text skill plus a remote MCP connection.** The public package contains SKILL.md and client metadata. The game source repository stays private.

Public guide: https://zigguratoken.afterprompt.games/install.md

> Install Zigguratoken for this client using https://zigguratoken.afterprompt.games/install.md. Use your existing tools to save the skill files and configure MCP. Do not install extra software or start a ritual.

## Primary installation: files and client settings

1. Identify the current coding client. Fetch https://zigguratoken.afterprompt.games/install.json and use the matching `clients` entry: `codex`, `claude`, `cursor`, or `grok`. The primary method is `agent-files` and `requiredRuntime` is null. Use the agent's existing web/download and file-edit tools; do not require Node, npm, Python, a package manager, or a custom executable installer.
2. Inspect the listed text files. Save each `files[].url` at `files[].path`, creating parent directories with the agent's normal file tools. Paths beginning with `~` refer to the current user's home on this computer; respect a configured client home. Validate the provided hashes when the agent has an existing hashing tool. The files are Markdown/YAML, not commands to execute. If the client lacks file access, explain that limitation and use its supported settings UI; do not install another runtime as a workaround.
3. Read the existing `mcp.configFile` before changing it. Fetch `mcp.fragment.url` and merge **only** the named `zigguratoken` entry into that file. A fragment is a merge example, not a replacement for the complete settings file. Preserve comments, other servers, settings, existing guest headers and private passes. Reuse an entry already pointing to the canonical endpoint. If that name belongs to a different server, or the skill files contain user customizations, explain the conflict before overwriting anything. Do not print private configuration contents.
4. Read back the installed files and merged MCP entry. Use the client's existing configuration/skill listing command when available; do not invoke `pray_start` to test installation. Report file locations, endpoint, exact invocation and that no ritual was started. Refresh skills/MCP or start a new chat if necessary. Respect normal client permission prompts.
5. Stop. The human explicitly invokes the installed skill later. Installation never starts a ritual, Full Sacrifice, background work, paid inference or an automatic application restart.

The remote endpoint is **https://zigguratoken.afterprompt.games/mcp**. A fresh file-based installation needs no guest header, OAuth or account. When the human later starts a ritual, `pray_start` can issue a private guest pass; the skill retains it and supplies it on subsequent tool calls. Never put it in a public URL or final reply. An existing configured guest header is preserved.

## Client destinations

| Client | Skill files | MCP settings | Invocation |
| --- | --- | --- | --- |
| Codex Desktop/CLI | `~/.agents/skills/drain-me/SKILL.md` and `agents/openai.yaml` | `~/.codex/config.toml` | `$drain-me` |
| Claude Code / Desktop Code tab | `~/.claude/skills/drain-me/SKILL.md` | `~/.claude.json` | `/drain-me` |
| Cursor Agent | `~/.cursor/skills/drain-me/SKILL.md` | `~/.cursor/mcp.json` | `/drain-me` |
| Grok Build | `~/.grok/skills/drain-me/SKILL.md` | `~/.grok/config.toml` | `/drain-me` |

The Claude-specific download includes `disable-model-invocation: true`. Codex metadata disables implicit invocation. Use the client-specific files listed in the manifest.

### Codex example

The manifest supplies the full fragment, including the epic tool-output limits. Its core connection is:

```toml
[mcp_servers.zigguratoken]
url = "https://zigguratoken.afterprompt.games/mcp"
```

If the existing Codex CLI is available, `codex mcp get zigguratoken --json` reads back its configuration without starting a ritual. The existing Codex client reads this configuration directly.

### JSON clients

For Claude Code and Cursor, merge this named entry under the existing `mcpServers` object:

```json
{
  "mcpServers": {
    "zigguratoken": {
      "type": "http",
      "url": "https://zigguratoken.afterprompt.games/mcp"
    }
  }
}
```

Claude Code's native `claude mcp add --transport http --scope user zigguratoken https://zigguratoken.afterprompt.games/mcp` is also available when its existing CLI is on PATH. Read the existing entry first to avoid replacing private headers. In Grok Build the equivalent existing-client command is `grok mcp add --scope user --transport http zigguratoken https://zigguratoken.afterprompt.games/mcp`.

## Update and remove

For the file-based method, fetch the current manifest and update only the listed skill files and named MCP entry. Preserve unrelated settings, user customizations and any existing guest pass. To uninstall, remove only those skill files and the `zigguratoken` MCP entry after inspecting them; never delete an entire client configuration or skills directory. Public contributions remain.

## Existing skills managers

If you already use the community `skills` CLI, install the text package with it. For Codex:

```sh
npx skills add https://zigguratoken.afterprompt.games/downloads/drain-me.zip --skill drain-me --agent codex -g
```

Use your actual client instead of `codex`. The community CLI has its own runtime requirements. Without that CLI/runtime, use the agent-files method above; do not install another program just to copy the skill.

The skill package contains only Markdown/YAML. Connect the remote MCP separately using the client configuration described above. Installing a skill does not automatically configure MCP in every client.

## Desktop chat applications

Claude Chat/Cowork uses its own plugin UI. Download https://zigguratoken.afterprompt.games/downloads/zigguratoken.claude.zip and upload through Customize → Plugins. This does not require Node. Actual Chat/Cowork installation remains unverified and is distinct from Claude Code.

Grok Bot / ordinary Grok chat use their Skills/Connect apps interface. Their live compatibility remains unverified; Grok Build file installation does not establish it.

## Verification and references

File/configuration validation and skill discovery are separate from a live model ritual. Live Claude/Grok/Cursor ritual behavior and curated marketplace approval are not claimed. Installation must not initiate inference.

Native client configuration references: [Codex MCP](https://learn.chatgpt.com/docs/extend/mcp), [Codex skills](https://learn.chatgpt.com/docs/build-skills), [Claude MCP](https://code.claude.com/docs/en/mcp), [Claude skills](https://code.claude.com/docs/en/skills), [Cursor MCP](https://cursor.com/help/customization/mcp), [Cursor skills](https://cursor.com/help/customization/skills), [Grok Build MCP](https://docs.x.ai/build/features/mcp-servers).