How it works

flowchart LR
    subgraph Input["Code Source"]
        Source1[GitHub Repository]
        Source2[Local Directory]
    end

    subgraph Pipeline["Salt Docs Pipeline"]
        Crawl[Crawl & Analyze]
        Identify[LLM Identify Abstractions]
        Generate[Generate Markdown Docs]
    end

    subgraph Output["Local Wiki"]
        Docs[Markdown Files]
        MCP[MCP Server]
    end

    subgraph Assistants["AI Assistants"]
        Cursor[Cursor]
        Claude[Claude]
        Continue[Continue]
    end

    Source1 --> Crawl
    Source2 --> Crawl
    Crawl --> Identify --> Generate --> Docs --> MCP
    MCP --> Cursor
    MCP --> Claude
    MCP --> Continue

Loading

Installation

Option 1: Install from PyPI

pip install salt-docs

Option 2: Install from source

git clone https://github.com/usesalt/salt-docs.git
cd salt-docs
pip install -e .

Quick Start

1. Initial Setup

Run the setup wizard to configure your API keys and preferences:

salt-docs init

2. Generate Documentation

Analyze GitHub repository

salt-docs run https://github.com/username/repo

Analyze local directory

salt-docs run /path/to/your/codebase

With custom options

salt-docs run https://github.com/username/repo --output /custom/path --language spanish --max-abstractions 10

Configuration

Salt Docs stores configuration in a per-user config file and uses your system's keyring for secure API key storage.

• macOS/Linux: ~/.config/saltdocs/config.json (or $XDG_CONFIG_HOME/saltdocs/config.json)
• Windows: %APPDATA%\saltdocs\config.json

Configuration Options

• llm_provider: LLM provider to use (gemini, openai, anthropic, openrouter, ollama) - default: gemini
• llm_model: Model name to use (e.g., "gemini-2.5-flash", "gpt-4o-mini", "claude-3-5-sonnet-20241022") - default: gemini-2.5-flash
• output_dir: Default output directory
• language: Default language for generated docs
• max_abstractions: Default number of abstractions to identify
• max_file_size: Maximum file size in bytes
• use_cache: Enable/disable LLM response caching
• include_patterns: Default file patterns to include
• exclude_patterns: Default file patterns to exclude
• ollama_base_url: Custom Ollama base URL (optional, default: http://localhost:11434)

Managing Configuration

View Current Configuration

salt-docs config show

Update API Keys

# Update API key for any provider (interactive)
salt-docs config update-api-key gemini
salt-docs config update-api-key openai
salt-docs config update-api-key anthropic
salt-docs config update-api-key openrouter

# Legacy command (still works, redirects to update-api-key)
salt-docs config update-gemini-key

# Update GitHub token (interactive)
salt-docs config update-github-token

# Update GitHub token directly
salt-docs config update-github-token "your-token-here"

Update Other Settings

# Change LLM provider
salt-docs config set llm-provider openai

# Change LLM model
salt-docs config set llm-model gpt-4o-mini

# Change default language
salt-docs config set language spanish

# Change max abstractions
salt-docs config set max_abstractions 15

# Disable caching
salt-docs config set use_cache false

# Update output directory
salt-docs config set output_dir /custom/path

---

MCP Server Setup

Salt Docs includes an MCP (Model Context Protocol) server that exposes your generated documentation to AI assistants in IDEs like Cursor, Continue.dev, and Claude Desktop.

MCP Tools Available

The MCP server provides these tools:

• list_docs - List all available documentation files
• get_docs - Fetch the full content of a documentation file (by resource name or absolute path)
• search_docs - Full-text search across documentation (paths, names, and resource names)
• index_directories - Index directories for fast searching

Setup Instructions

Cursor

1. Open or create your MCP configuration file:

   • macOS/Linux: ~/.cursor/mcp.json
   • Windows: %APPDATA%\Cursor\mcp.json

2. Add the salt-docs server configuration:

{
  "mcpServers": {
    "salt-docs": {
      "command": "salt-docs",
      "args": ["mcp"]
    }
  }
}

3. Restart Cursor to load the MCP server.

4. The AI assistant in Cursor can now access your documentation using tools like:

   • "What documentation do we have?"
   • "Get me the documentation for 'SALT project"
   • "Read the README documentation"

Claude Desktop

1. Open or create your Claude configuration file:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
   • Linux: ~/.config/Claude/claude_desktop_config.json

2. Add the salt-docs server configuration:

{
  "mcpServers": {
    "salt-docs": {
      "command": "salt-docs",
      "args": ["mcp"]
    }
  }
}

3. Restart Claude Desktop to load the MCP server.

Troubleshooting

• Command not found: Make sure salt-docs is in your PATH. You can verify by running salt-docs --version in your terminal.
• Server not starting: Ensure you've run salt-docs init and have generated at least one documentation project.
• No docs found: The MCP server discovers docs from your configured output_dir. Run salt-docs config show to check your output directory.

Testing the MCP Server

You can test the MCP server directly:

salt-docs mcp

This will start the server in stdio mode (for MCP clients). To test locally, you can use the test scripts in the tests/ directory.

LLM Provider Support

Salt Docs supports multiple LLM providers, allowing you to choose the best option for your needs:

Supported Providers

1. Google Gemini (default)

   • Recommended models: gemini-2.5-pro, gemini-2.5-flash, gemini-1.5-pro, gemini-1.5-flash
   • API key required: Yes (GEMINI_API_KEY)

2. OpenAI

   • Recommended models: gpt-4o-mini, gpt-4.1-mini, gpt-5-mini, gpt-5-nano
   • API key required: Yes (OPENAI_API_KEY)
   • Supports o1 models with reasoning capabilities

3. Anthropic Claude

   • Recommended models: claude-3-5-sonnet, claude-3-5-haiku, claude-3-7-sonnet (with extended thinking), claude-3-opus
   • API key required: Yes (ANTHROPIC_API_KEY)

4. OpenRouter

   • Recommended models: google/gemini-2.5-flash:free, meta-llama/llama-3.1-8b-instruct:free, openai/gpt-4o-mini, anthropic/claude-3.5-sonnet
   • API key required: Yes (OPENROUTER_API_KEY)
   • Access multiple models through a single API

5. Ollama (Local)

   • Recommended models: llama3.2, llama3.1, mistral, codellama, phi3
   • API key required: No (runs locally)
   • Default URL: http://localhost:11434
   • Perfect for privacy-sensitive projects or offline usage

Switching Providers

You can switch between providers at any time:

# Switch to OpenAI
salt-docs config set llm-provider openai
salt-docs config set llm-model gpt-4o-mini
salt-docs config update-api-key openai

# Switch to Ollama (local)
salt-docs config set llm-provider ollama
salt-docs config set llm-model llama3.2
# No API key needed for Ollama!

CLI Options

Required

• run - GitHub repo URL, current open directory or local directory path
• --repo or --dir - GitHub repo URL or local directory path (depricated)

Optional

• -n, --name - Project name (derived from repo/directory if omitted)
• -t, --token - GitHub personal access token
• -o, --output - Output directory (overrides config default)
• -i, --include - File patterns to include (e.g., ".py", ".js")
• -e, --exclude - File patterns to exclude (e.g., "tests/", "docs/")
• -s, --max-size - Maximum file size in bytes (default: 100KB)
• --language - Language for generated docs (default: "english")
• --no-cache - Disable LLM response caching
• --max-abstractions - Maximum number of abstractions to identify (default: 10)