INSTALLGUIDE.

One-line install. Skill registers as a project-scoped MCP server, exposes its tool to Claude Code, called on any prompt that matches the skill description. No restart needed.

1. Unzip the skill bundle into your project (or anywhere on disk you can reference).
2. From your project root, run the command below. Claude Code adds the server to .mcp.json and reloads automatically.
3. Type /mcp in any Claude Code session to confirm the server is connected.

claude mcp add your-skill --command node --args ./your-skill/mcp/server.js

Success state: /mcp lists your-skill as connected.

Paste a server entry into your Claude Desktop config and the skill is available in every chat. Claude calls the tool when relevant.

1. Open claude_desktop_config.json. macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. Windows: %APPDATA%\Claude\claude_desktop_config.json.
2. Add the snippet below under mcpServers. If the file is empty, paste the full block.
3. Quit and reopen Claude Desktop. The server appears in the tools tray at the bottom of the input.

{
  "mcpServers": {
    "your-skill": {
      "command": "node",
      "args": ["/absolute/path/to/your-skill/mcp/server.js"]
    }
  }
}

Success state: The skill appears in the Claude Desktop tools tray.

Claude.ai Projects don't run MCP servers, so the skill installs as a Project: SKILL.md goes in Project Instructions, knowledge files go in Project Knowledge. Behaves identically to the MCP version, just invoked by typing the trigger phrase.

1. Create a new Project on claude.ai (or open an existing one).
2. Open Project Instructions. Paste the full contents of SKILL.md from the skill bundle.
3. Open Project Knowledge. Upload every file from the skill's /knowledge/ folder.
4. Open a chat in the Project. Type the trigger phrase listed at the top of the per-skill page.

Success state: SKILL.md fills Project Instructions, knowledge files are uploaded, the trigger phrase produces the expected output.

Claude Cowork (Claude for Work team workspace) installs the skill as workspace instructions. Once installed, every member of the workspace can invoke the skill from any new chat.

1. Open your Cowork workspace. Go to Settings, then Workspace instructions.
2. Paste the full SKILL.md into the Workspace instructions field.
3. Upload the knowledge files to the workspace shared file library (or to a designated Project inside the workspace).
4. Confirm the skill works by typing the trigger phrase in a new chat.

Note: If your team works mostly inside Projects, install the skill at the Project level (see Claude.ai Projects above) instead of the whole workspace. Project-scoped installs keep workspace instructions clean.

Build a Custom GPT that wraps the skill. Instructions hold SKILL.md, Knowledge holds the corpus, optional MCP Action exposes the same server if the skill supports remote MCP. Best path if you want the skill in your team GPT picker.

1. Open ChatGPT, go to Explore GPTs, then Create.
2. In the Configure tab, paste the full SKILL.md into Instructions.
3. Upload every file from the skill /knowledge/ folder into Knowledge.
4. Optional: in Actions, add the skill hosted MCP endpoint if one is published.
5. Save. Open a chat with your new GPT and type the trigger phrase.

If your ChatGPT plan supports remote MCP servers (Team / Enterprise / Pro with the connector enabled), point ChatGPT at the skill hosted MCP endpoint. Same brain as Claude Code, called natively from ChatGPT.

1. In ChatGPT, open Settings, then Connectors.
2. Click Add MCP Server. Paste the skill hosted MCP endpoint (listed on the per-skill page if available). For local-only skills, run the MCP server on your machine and expose it via the bundled tunnel script.
3. Authorize the connector when prompted.
4. Open a new chat. ChatGPT auto-routes prompts that match the skill tool description.

Name: your-skill
URL: https://your-skill.example.com/mcp
Auth: (skill bundle ships an API key if required)

Note: Not every skill ships a hosted endpoint. If yours does not, use the Custom GPT path or the SKILL.md fallback. The per-skill page tells you which paths are supported.

Gemini Code Assist and Gemini Advanced both support MCP via Extensions. Add the skill to your Gemini extensions config and the tool is callable in any conversation.

1. Open your Gemini extensions config. For Gemini Code Assist this lives in your IDE settings. For Gemini Advanced (web), open Extensions, then Manage MCP servers.
2. Add the snippet below. Adjust the absolute path to point at your unzipped skill bundle.
3. Reload Gemini (or restart the IDE). The skill tool name appears in the extensions list.
4. Type the trigger phrase listed on the per-skill page.

{
  "extensions": {
    "your-skill": {
      "command": "node",
      "args": ["/absolute/path/to/your-skill/mcp/server.js"]
    }
  }
}

Every Made With Machines skill ships a single SKILL.md file that holds the entire brain. Paste it into any LLM that accepts long-form instructions and the skill works. No MCP, no setup, no config files.

1. Open SKILL.md from the skill bundle.
2. Copy the full contents.
3. Paste into your LLM system prompt, instructions field, or first user message.
4. If the skill needs reference files (corpus, templates), paste them inline below SKILL.md or attach them however the tool allows.
5. Type the trigger phrase. The skill runs.

Note: Why this works: every MWM skill is built so SKILL.md is the source of truth. The MCP server is a thin wrapper that loads SKILL.md and exposes one tool. Paste-in installs use the same brain, just without the tool-call layer. Output is identical.