Skip to content

INSTALL.md

Latest commit

 

History

History
239 lines (177 loc) · 5.44 KB

INSTALL.md

File metadata and controls

239 lines (177 loc) · 5.44 KB

SLANG — Installation Guide

SLANG ships as both a CLI tool and an MCP server, so it plugs into every major AI environment without friction. No code needed to get started — write your workflow, your LLM runs it.

Quick install (npm)

npm install -g @riktar/slang

This installs two binaries:

Binary Purpose
slang CLI: run, parse, check, prompt
slang-mcp MCP server over stdio

Environment Variables

All adapters can be configured through environment variables instead of passing flags each time:

Variable Description
SLANG_ADAPTER sampling (default) | openai | anthropic | echo. When running inside an MCP host (Claude Code, Claude Desktop) the default sampling delegates LLM calls to the host — no API key required.
SLANG_API_KEY API key for openai/anthropic adapters (falls back to OPENAI_API_KEY / ANTHROPIC_API_KEY). Not required for sampling.
SLANG_MODEL Model override (e.g. gpt-4o, claude-opus-4-5)
SLANG_BASE_URL Custom base URL for OpenAI-compatible endpoints

Claude Code

Add SLANG as an MCP server so Claude Code can run, parse, and check .slang files directly.

claude mcp add slang -- npx --package @riktar/slang slang-mcp

No API key needed — SLANG defaults to the sampling adapter, which delegates LLM calls back to Claude through the MCP protocol using your existing Claude subscription.

If you prefer to use a separate API key (e.g. to charge to a different account):

claude mcp add slang -- env SLANG_ADAPTER=anthropic SLANG_API_KEY=sk-ant-... npx --package @riktar/slang slang-mcp

Verify it's registered:

claude mcp list

Claude Code will now have four tools: run_flow, parse_flow, check_flow, get_zero_setup_prompt.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or
%APPDATA%\Claude\claude_desktop_config.json (Windows):

Zero-key setup (recommended) — uses your Claude subscription:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp"
    }
  }
}

SLANG defaults to the sampling adapter, which delegates all LLM calls back to Claude Desktop through the MCP protocol. No API key needed.

With a separate Anthropic key:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp",
      "env": {
        "SLANG_ADAPTER": "anthropic",
        "SLANG_API_KEY": "sk-ant-YOUR_KEY_HERE"
      }
    }
  }
}

With an OpenAI key:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp",
      "env": {
        "SLANG_ADAPTER": "openai",
        "SLANG_API_KEY": "sk-YOUR_OPENAI_KEY"
      }
    }
  }
}

OpenAI Desktop (ChatGPT)

OpenAI Desktop supports MCP servers through its settings:

  1. Open ChatGPT → Settings → Connected apps → Add MCP server
  2. Fill in:
    • Name: slang
    • Command: slang-mcp
  3. Click Save and reload the app.

SLANG will use the sampling adapter by default, which asks the ChatGPT host to run the LLM calls. If you want to force a specific model or use your own key, add environment variables:

SLANG_ADAPTER=openai
SLANG_API_KEY=sk-YOUR_KEY

Any MCP-compatible host (generic config)

Any host that accepts the standard MCP JSON config block. Minimal config — no API key needed thanks to the sampling default:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp"
    }
  }
}

With an explicit adapter and key:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp",
      "args": [],
      "env": {
        "SLANG_ADAPTER": "openai",
        "SLANG_API_KEY": "sk-..."
      }
    }
  }
}

For hosts that require an absolute path (e.g. some Docker setups):

{
  "command": "node",
  "args": ["/usr/local/lib/node_modules/@riktar/slang/dist/mcp.js"]
}

Local / Ollama (zero-cost)

Point SLANG at a local OpenAI-compatible endpoint:

{
  "mcpServers": {
    "slang": {
      "command": "slang-mcp",
      "env": {
        "SLANG_ADAPTER": "openai",
        "SLANG_API_KEY": "ollama",
        "SLANG_BASE_URL": "http://localhost:11434/v1",
        "SLANG_MODEL": "llama3.2"
      }
    }
  }
}

Zero-setup (no runtime, any LLM)

No install required. Get the interpreter prompt and paste it into any LLM's system prompt:

slang prompt

Or via MCP:

Tool: get_zero_setup_prompt

Then paste the output as the system prompt in ChatGPT, Claude.ai, Gemini, etc. — the LLM becomes a SLANG interpreter and can execute any .slang flow pasted in the chat.

Available MCP Tools

Tool Description
run_flow Execute a SLANG flow; returns final state, agent outputs, status
parse_flow Parse source to AST JSON; validates syntax
check_flow Dependency graph analysis + deadlock detection
get_zero_setup_prompt Returns the zero-setup system prompt for paste-into-LLM use

CLI Reference

slang run <file.slang>        # execute a flow
slang parse <file.slang>      # dump AST as JSON
slang check <file.slang>      # dependency + deadlock report
slang prompt                  # print zero-setup system prompt

# Adapter flags
--adapter openai|anthropic|echo   # (MCP mode defaults to 'sampling')
--api-key sk-...                  # not required with MCP sampling
--model gpt-4o
--base-url http://localhost:11434/v1