Markdown Navigation MCP
An MCP server that provides efficient navigation and reading of large markdown files using ctags to reduce context usage
Markdown Navigation MCP Server
Efficiently navigate large markdown files (2,000+ lines) without loading entire documents into context. Reduces token usage by 50-80% when working with documentation, planning files, and technical specifications.
Quick Start
Prerequisites: Universal Ctags and Go 1.21+
# Install ctags
brew install universal-ctags # macOS
sudo apt install universal-ctags # Ubuntu/Debian
sudo dnf install universal-ctags # Fedora
# Build and install
git clone <repo-url>
cd markdown-mcp
go build -o mdnav-server ./cmd/server
sudo cp mdnav-server /usr/local/bin/
Configure Claude Code (~/claude.json):
{
"mcpServers": {
"markdown-nav": {
"command": "mdnav-server"
}
}
}
Features
- Zero-configuration: Automatic ctags execution on-demand
- Smart caching: Sub-microsecond responses for repeated queries
- Auto-invalidation: Cache updates when files change
- Selective reading: Load only the sections you need
- Tree navigation: View document structure without reading content
- Pattern matching: Find sections by regex patterns
- Depth control: Limit tree/section depth for focused views
Tools
markdown_tree
Display document structure as tree (ASCII or JSON format).
Key parameters:
file_path: Path to markdown fileformat: "ascii" or "json" (default: "json")max_depth: Limit tree depth 1-6 (default: 2 shows H1+H2)section_name_pattern: Regex to filter sections
markdown_section_bounds
Get line number boundaries for a specific section.
Key parameters:
file_path: Path to markdown filesection_heading: Exact heading text (without # symbols)
markdown_read_section
Read content from a specific section.
Key parameters:
file_path: Path to markdown filesection_heading: Exact heading text (without # symbols)max_subsection_levels: Limit subsection depth (omit for all)
markdown_list_sections
List all sections with filters.
Key parameters:
file_path: Path to markdown filemax_depth: Maximum heading level to show (default: 2)section_name_pattern: Regex to filter section names
Usage Examples
Finding and reading a specific task
User: "Review Task 4 from the planning document"
Claude uses:
1. markdown_tree to see document structure
2. markdown_section_bounds to find Task 4 location
3. markdown_read_section to read only Task 4 content
Result: Complete task analysis using only relevant section (~200 lines instead of 2000)
Discovering documentation sections
User: "What testing strategies are documented?"
Claude uses:
1. markdown_list_sections with pattern="test" to find testing sections
2. markdown_read_section for each relevant section
Result: Comprehensive overview without loading entire document
For more detailed tool usage examples with real output, see examples/EXAMPLES.md.
Configuration
Custom ctags path
If ctags is not in PATH, specify location:
{
"mcpServers": {
"markdown-nav": {
"command": "mdnav-server",
"args": ["-ctags-path", "/custom/path/to/ctags"]
}
}
}
Troubleshooting
"ctags not found in PATH"
- Install Universal Ctags or use
-ctags-pathflag
"section not found"
- Use exact heading text (case-sensitive, no # symbols)
- Run
markdown_list_sectionsto see available sections
"no entries found"
- Ensure file has markdown headings (#, ##, ###, ####)
- Verify Universal Ctags (not Exuberant) is installed
Cache issues
- Restart MCP server to clear cache (auto-invalidates on file changes)
Development
For implementation details, architecture, and contributing guidelines, see CLAUDE.md.
Quick development commands:
go test ./... # Run tests
golangci-lint run # Lint code
go build ./cmd/server # Build server
License
[Your license here]
Related Servers
Logfire
Provides access to OpenTelemetry traces and metrics through Logfire.
MCP Server Creator
A meta-server for dynamically generating MCP server configurations and Python code.
Image
Fetch and process images from URLs, local file paths, and numpy arrays, returning them as base64-encoded strings.
Argo CD
Interact with Argo CD applications through natural language.
Claude Code MCP
Orchestrates multiple Claude Code agents across iTerm2 sessions, providing centralized management and inter-agent communication.
Python Notebook MCP
Enables AI assistants to interact with local Jupyter notebooks (.ipynb).
MCP Remote
A remote proxy for MCP that enables local clients to connect to remote servers via OAuth.
Remote MCP Server Authless
An example of a remote MCP server deployable on Cloudflare Workers without authentication.
Animated video MCP Server
Executes Manim Python animation scripts to generate and return videos.
FAL FLUX.1 Kontext [Max]
A frontier image generation and editing model with advanced text rendering and contextual understanding, powered by the FAL AI API.