Skytells MCP

Model Context Protocol server for the Skytells CLI, public model discovery, generated media downloads, and Skytells Learn.

Get Started

1. Create a Skytells account.
2. Open the Skytells Console to manage projects and API keys.
3. Browse Skytells Learn for guides and tutorials.
4. Read the Skytells API documentation.

Hosted Server

The public MCP server is available at:

https://mcp.skytells.ai

It provides public model discovery and Skytells Learn search without authentication.

Claude Code

claude mcp add --transport http skytells https://mcp.skytells.ai

Gemini CLI

gemini mcp add --transport http skytells https://mcp.skytells.ai

VS Code and GitHub Copilot

{
  "servers": {
    "skytells": {
      "type": "http",
      "url": "https://mcp.skytells.ai"
    }
  }
}

Cursor

{
  "mcpServers": {
    "skytells": {
      "url": "https://mcp.skytells.ai"
    }
  }
}

Local Installation

Requires Node.js 20 or newer. The MCP can install the CLI through skytells_cli_install, or install it directly:

npm install -g @skytells/cli

Run the MCP directly from npm without a global installation:

npx -y @skytells/mcp

npx downloads the published npm package and starts its stdio MCP server. The GitHub repository can remain private.

VS Code and GitHub Copilot

Add to .vscode/mcp.json or your user MCP configuration:

{
  "servers": {
    "skytells": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@skytells/mcp"]
    }
  }
}

Claude Code

claude mcp add --transport stdio --scope user skytells -- npx -y @skytells/mcp

Cursor

Add to ~/.cursor/mcp.json or .cursor/mcp.json:

{
  "mcpServers": {
    "skytells": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@skytells/mcp"]
    }
  }
}

Gemini CLI

gemini mcp add --scope user skytells npx -y @skytells/mcp

OpenAI Codex CLI

codex mcp add skytells -- npx -y @skytells/mcp

Authentication

The MCP never returns credential values. The Skytells CLI supports three independent credentials:

| Operations | Credential | | --- | --- | | Account and user projects | skytells login or SKYTELLS_TOKEN | | Linked-project resources | skytells link <key> or SKYTELLS_ACCESS_KEY | | Models and predictions | skytells api-key set or SKYTELLS_API_KEY |

Models and predictions require the separate Skytells API key. Interactive login and API-key entry must run in the user's terminal; skytells_auth_setup returns the correct command and console URL.

Tools

Public tools:

• list_models: list/filter the public catalog.
• search_models: search by name, namespace, description, vendor, modality, or capability.
• get_model: fetch {vendor}/{model}.
• skytells_learn_search: search Learn, optionally scoped by tag such as api.

Local tools additionally include CLI status/install, authentication status/setup, account/projects, prediction create/get/list/cancel/delete, save_output_to_file, and save_generated_outputs.

Complete CLI Coverage

skytells_run covers every non-interactive command in the installed CLI, including projects, project settings, apps, deployments, Drops, models, predictions, databases, environment variables, domains, members, status, logs, cloud infrastructure, workflows, Orchestrator, Cloud Agents, and Cognition. Destructive commands require explicit confirmation.

Use skytells_cli_help with a command path, such as {"command":["agents","runs"]}, to read live help from the installed CLI. This keeps coverage current as Skytells/cli adds commands. Interactive login, api-key set, and indefinite logs --follow are terminal workflows rather than bounded MCP calls.

For generation, call get_model first. Its input_schema defines the exact object for skytells_predictions_create; pass the returned bare namespace as model (for example, mera-avatar, not skytells/mera-avatar).

License

MIT