DocsetMCP

- 📚Instant Documentation: No switching, no web searches. Get straight to the docs directly in your AI conversation
- 🔒Local and Private: Work with docset files on your machine
- ⚡Lightning Fast: Optimized caching and direct database queries
- 🎯Precise Results: Get exactly what you need with smart filtering

By default, DocsetMCP looks for docsets in Dash's standard directories:

- Docsets:~/Library/Application Support/Dash/DocSets
- Cheatsheets:~/Library/Application Support/Dash/Cheat Sheets

You can customize these locations using:

# Set custom docset directory export DOCSET_PATH="/path/to/your/docsets" # Set custom cheatsheet directory export CHEATSHEET_PATH="/path/to/your/cheatsheets" # Run with custom paths docsetmcp

# Test with custom docset path docsetmcp --docset-path "/path/to/your/docsets" --list-docsets # Test with custom cheatsheet path docsetmcp --cheatsheet-path "/path/to/your/cheatsheets" --test-connection # Use both custom paths docsetmcp --docset-path "/custom/docsets" --cheatsheet-path "/custom/cheatsheets" # Use additional search paths (searches multiple locations) docsetmcp --additional-docset-paths "/extra/docsets" "/more/docsets" docsetmcp --additional-cheatsheet-paths "/extra/cheatsheets" "/more/cheatsheets"

- CLI arguments (highest priority)
- Environment variables
- Default Dash locations (lowest priority)

The--additional-docset-pathsand--additional-cheatsheet-pathsoptions allow DocsetMCP to search in multiple locations beyond the primary path. This is useful when:

- You have docsets in multiple directories
- You want to include third-party or custom docsets
- You're sharing docsets across different tools

DocsetMCP will automatically discover and configure docsets found in these additional paths.

Choose your MCP client below for specific setup instructions:

Add to~/Library/Application Support/Claude/claude_desktop_config.json:

If you prefer to install locally or your MCP client doesn't supportuvx:

Then usedocsetmcpinstead ofuvx docsetmcpin your configuration.

git clone https://github.com/codybrom/docsetmcp.git cd docsetmcp pip install -e .

# Install test dependencies pip install pytest pytest-cov pytest-xdist # Run basic tests pytest tests/test_docsets.py::TestDocsets::test_yaml_structure -v # Run quick tests (structure + existence checks) pytest tests/ -k "yaml_structure or test_docset_exists" -v # Run full test suite (all docsets) pytest tests/ -v # Run with coverage pytest tests/ --cov=docsetmcp --cov-report=html -v # Validate all local cheatsheets work (integration test) python scripts/validate_cheatsheets.py

Once configured, you can ask your AI assistant to search documentation naturally:

"Search for URLSession documentation" "Show me how to use AppIntent in SwiftUI" "Find CarPlay framework documentation" # Returns framework + related entries with drilldown notes "Search for CPListTemplate class" # Returns specific CarPlay class "Find NSPredicate examples"

"Look up Express.js middleware documentation" "Search React hooks in the React docset" "Find CSS flexbox properties"

"Search git rebase commands in the Git cheatsheet" "Show Docker compose syntax from the cheatsheet" "Find bash array manipulation commands"

"Search pandas DataFrame methods" "Look up NumPy array broadcasting" "Find matplotlib pyplot functions"

# Search specific docset with language filter "Use search_docs for 'URLSession' in the apple_api_reference docset with Swift language" # Explore framework members using drilldown guidance "Search for 'SwiftData' then follow the drilldown note to see all members" # List all available tools "What frameworks are available in the nodejs docset?" # Browse cheatsheet categories "Show all categories in the vim cheatsheet"

DocsetMCP is designed forname-based searches, not keyword searching. Follow this workflow:

# Find what languages are available "List all available programming languages" # Find docsets for your language "Show me all Python docsets" # See what types are available in a docset "List all types in the apple_api_reference docset for Swift" # Browse entries by type with letter filters "Show me all Classes starting with 'UI' in apple_api_reference for Swift"

# Once you know exact names, search for them "Search for UIViewController in apple_api_reference with Swift" "Find readFile documentation in nodejs docset" "Show me the CarPlay framework documentation"

When you find container types (frameworks, classes), follow the drilldown guidance:

# Container entry will show: "contains 42 additional members - use search_docs('ContainerName', max_results=50)" "Search for SwiftData in apple_api_reference with max_results=50"

- Multi-Format Support: Handles both Apple cache format and tarix compression
- Direct Database Access: Queries Dash's SQLite databases for fast lookups
- Name-Based Matching: Only returns entries where search terms match item names (no false positives)
- Smart Ranking: Prioritizes exact matches, then prefix matches, then substring matches
- Dynamic Type Ordering: Uses docset configuration files for intelligent result prioritization
- Container Detection: Automatically detects frameworks/classes with members and provides exploration guidance
- Smart Extraction: Decompresses Apple's DocC JSON or extracts HTML from tarix archives
- Markdown Formatting: Converts documentation to readable Markdown

DocsetMCP provides eleven powerful tools for accessing your documentation:

Search and extract documentation from any docset.

Search Dash cheatsheets for quick command reference.

List all installed Dash docsets with their supported languages.

List all available Dash cheatsheets that can be searched.

List frameworks/types within a specific docset.

Discover all programming languages with available documentation.

Find all docsets that support a specific programming language.

List all available types (Class, Protocol, Function, etc.) in a docset/language.

List entries filtered by type and optional name prefix.

List all categories within a specific cheatsheet.

Fetch entire cheatsheet content (recommended for comprehensive access).

This means the docset isn't installed in Dash. To fix:
- Open Dash.app
- Go to Preferences → Downloads
- Download the required docset
- Restart your MCP client
- Check installation: Runpip show docsetmcpto verify installation
- Test manually: Runuvx docsetmcpin terminal - you should see MCP output
- Check logs:

- Claude Desktop: Check Console.app for Claude logs
- Cursor: Check Output → MCP panel

# Clone the repository git clone https://github.com/codybrom/docsetmcp.git cd docsetmcp # Install in development mode pip install -e . # Install all development dependencies pip install -r requirements.txt # Set up pre-commit hooks pre-commit install

# Run basic structure tests pytest tests/test_docsets.py::TestDocsets::test_yaml_structure -v # Run quick tests (structure + existence) pytest tests/ -k "yaml_structure or test_docset_exists" -v # Run full test suite (all docsets) pytest tests/ -v # Run with coverage pytest tests/ --cov=docsetmcp --cov-report=html -v # Run tests in parallel pytest tests/ -n auto -v # Validate cheatsheets python scripts/validate_cheatsheets.py

# Format Python code with Black black docsetmcp/ # Format YAML files with yamlfix yamlfix docsetmcp/docsets/*.yaml # Run all pre-commit hooks pre-commit run --all-files # Run specific hook pre-commit run yamlfix --all-files # Run spell check (cspell installed automatically during setup) npm run spell

# Test version docsetmcp --version # List available docsets docsetmcp --list-docsets # Test server startup docsetmcp --test-connection # Test with custom paths docsetmcp --docset-path "/custom/path" --list-docsets

# Build package python setup.py sdist bdist_wheel # Install from source pip install .

-

docsetmcp/server.py: Main MCP server implementation using FastMCP. Contains the DashExtractor class that handles:

- Apple cache format (SHA-1 UUID-based with brotli compression)
- Tarix format (tar.gz archives)
- SQLite database queries for documentation lookup
- HTML to Markdown conversion

docsetmcp/config_loader.py: Configuration system that loads YAML configs for 165+ supported docsets. Provides smart defaults and handles both simple and complex configuration formats.

docsetmcp/docsets/: YAML configuration files for each supported docset, defining:

- Docset paths and formats
- Language variants and filters
- Type priorities for search results
-

Multi-Format Support: The server detects and handles both Apple's modern cache format (using SHA-1 based UUIDs) and the older tarix compression format automatically based on docset configuration.

Caching Strategy: Extracted documentation is cached in memory (_fs_cache for Apple format, _html_cache for tarix) to improve performance on repeated queries.