@almyty/mcp-server

Local MCP server that connects your AI coding agent to almyty tools and skills.

Quick Start

npx @almyty/mcp-server

Setup

Claude Code

claude mcp add almyty -- npx -y @almyty/mcp-server

Cursor / Windsurf (.cursor/mcp.json)

{
  "mcpServers": {
    "almyty": {
      "command": "npx",
      "args": ["-y", "@almyty/mcp-server"]
    }
  }
}

VS Code Copilot (.vscode/mcp.json)

{
  "servers": {
    "almyty": {
      "command": "npx",
      "args": ["-y", "@almyty/mcp-server"]
    }
  }
}

Google Gemini CLI (~/.gemini/settings.json)

{
  "mcpServers": {
    "almyty": {
      "command": "npx",
      "args": ["-y", "@almyty/mcp-server"]
    }
  }
}

OpenAI Codex CLI (~/.codex/config.toml)

[mcp_servers.almyty]
command = "npx"
args = ["-y", "@almyty/mcp-server"]

How It Works

Instead of dumping every tool schema into the LLM context (thousands of tokens), this server uses a skill-first approach:

1. 2 tools exposed to the LLM: almyty_execute + almyty_search
2. Skills as prompts loaded on demand when the LLM needs them
3. LLM searches for relevant tools → loads the skill → calls execute

Traditional MCP: 20 tools × ~200 tokens = ~4,000 tokens (always in context)
Skill-first:     2 tools + on-demand skills = ~300 tokens base

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | ALMYTY_URL | https://api.almyty.com | almyty API URL | | ALMYTY_TOKEN | — | Auth token (or use npx @almyty/mcp-server login) | | ALMYTY_GATEWAY_ID | — | Scope to a specific gateway |

Authentication

# Interactive login (opens browser)
npx @almyty/mcp-server login

# Or set token directly
export ALMYTY_TOKEN=your-token
npx @almyty/mcp-server

License

BSL-1.1