Quillmark MCP Server

A Model Context Protocol (MCP) server that exposes Quillmark's document rendering capabilities to AI models and MCP clients.

Overview

The Quillmark MCP server enables AI assistants to:

• Discover available document templates (Quills)
• Understand template requirements and frontmatter schemas
• Validate document frontmatter before rendering
• Render markdown documents to PDF or SVG formats
• Receive rich error diagnostics to help users fix issues

Installation

Run with uvx (recommended):

uvx quillmark-mcp

# Using uv (recommended)
uv pip install quillmark-mcp

# Using pip
pip install quillmark-mcp

Quick Start

Running the Server

# Run using Python module
python -m quillmark_mcp

# Or using uv
uvx quillmark-mcp

MCP Client Configuration

Add to your MCP client configuration (e.g., Claude Desktop):

{
  "mcpServers": {
    "quillmark": {
      "command": "python",
      "args": ["-m", "quillmark_mcp"]
    }
  }
}

Available Tools

1. list_quills

List all registered Quill templates with metadata.

{}

Returns:

{
  "quills": [
    {
      "name": "taro",
      "backend": "typst",
      "description": "",
      "tags": []
    },
    {
      "name": "usaf_memo",
      "backend": "typst",
      "description": "",
      "tags": []
    }
  ]
}

2. get_quill_info

Get detailed information about a specific Quill template.

{
  "quill_name": "taro"
}

Returns frontmatter schema, required fields, and the example markdown content.

3. get_markdown_template

Get markdown template content by template name from the templates collection.

{
  "template_name": "U.S. Air Force Memo"
}

Returns the markdown content of the specified template from list_markdown_templates.

4. list_markdown_templates

List available markdown templates from the templates.json manifest.

{}

Returns:

{
  "templates": [
    {
      "name": "U.S. Air Force Memo",
      "description": "AFH 33-337 compliant official memorandum for the U.S. Air Force."
    },
    {
      "name": "U.S. Space Force Memo",
      "description": "Official memorandum template for the U.S. Space Force."
    }
  ]
}

5. render_document

Render a markdown document to PDF or SVG.

{
  "markdown": "---\nQUILL: taro\nauthor: John\nice_cream: Taro\ntitle: My Favorite\n...",
  "output_format": "PDF"
}

The quill_name parameter is optional if the markdown contains a QUILL: directive in the frontmatter.

6. validate_frontmatter

Validate frontmatter without rendering.

{
  "markdown": "---\nQUILL: taro\nauthor: John\n..."
}

Returns validation status and any missing required fields.

7. save_rendered_file

Save a rendered document from cache to a local file. Use this after render_document to save the file to disk.

{
  "resource_uri": "quillmark://render/a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "file_path": "/path/to/output.pdf"
}

Returns success status and file information:

{
  "success": true,
  "file_path": "/absolute/path/to/output.pdf",
  "file_size": 12345,
  "mime_type": "application/pdf",
  "format": "PDF"
}

Note: This tool is particularly useful when running the MCP server locally, allowing Claude Code and other clients to save rendered documents directly to your filesystem.

Usage Examples

AI-Assisted Document Writing

User: "Help me create a document using the taro template"

AI Workflow:

1. Call list_quills() to discover available quill templates
2. Call get_quill_info("taro") to learn requirements and get the example markdown
3. Generate markdown based on user's input
4. Call validate_frontmatter() to check correctness
5. Call render_document() to produce the final PDF (returns base64 data + resource_uri)
6. Call save_rendered_file() with the resource_uri to save the PDF to the user's filesystem

Note: For general-purpose templates like "U.S. Air Force Memo", use:

1. Call list_markdown_templates() to discover available templates
2. Call get_markdown_template("U.S. Air Force Memo") to get the template content
3. Modify the template for the specific use case
4. Call render_document() to produce the final PDF

Error Handling

The server provides rich diagnostics when errors occur:

{
  "success": false,
  "error_type": "ValidationError",
  "error_message": "Required fields missing",
  "diagnostics": [
    {
      "severity": "ERROR",
      "message": "Required field 'recipient' is missing",
      "code": "missing_field",
      "hint": "Add 'recipient: <value>' to your frontmatter"
    }
  ]
}

Extended YAML Metadata Standard

Quillmark supports structured content with the Extended YAML Metadata Standard:

• QUILL key: Specifies which template to use
• SCOPE key: Creates collections of content blocks

Example:

---
title: Product Catalog
---

Main description.

---
SCOPE: products
name: Widget
price: 19.99
---

Widget description.

---
SCOPE: products
name: Gadget
price: 29.99
---

Gadget description.

Architecture

┌──────────────────────────────┐
│       AI Model / Client      │
│    (via MCP protocol)        │
└──────────────┬───────────────┘
               │ MCP JSON-RPC
               ▼
┌──────────────────────────────┐
│     Quillmark MCP Server     │
│  - list_quills               │
│  - get_quill_info            │
│  - get_quill_template        │
│  - render_document           │
│  - validate_frontmatter      │
└──────────────┬───────────────┘
               │
               ▼
┌──────────────────────────────┐
│    Quillmark Engine          │
│  (Python package)            │
└──────────────────────────────┘

Design Principles

1. MCP-Native: Follows MCP protocol specifications
2. Rich Diagnostics: Provides helpful error messages with hints
3. Stateless: Each tool call is independent
4. JSON-Based: All data exchanged via JSON for compatibility

Development

Setup

# Clone the repository
git clone https://github.com/nibsbin/quillmark-mcp.git
cd quillmark-mcp

# Install dependencies
uv pip install -e ".[dev]"

Testing

# Run tests
pytest

# Run type checking
mypy src/quillmark_mcp

# Run linting
ruff check src/quillmark_mcp

Security Considerations

• Input Validation: Markdown size limits, YAML depth limits
• Output Safety: Binary outputs are base64-encoded
• Read-Only: All MCP tools are read-only operations
• Sandboxing: Consider running rendering in isolated environment

License

See LICENSE file for details.

References

• MCP Specification
• Quillmark Documentation
• Design Document
• Saving Files Locally - Guide for saving rendered documents to your filesystem

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

Status

Version: 0.1.0
Status: Implementation Phase
Python: 3.10+
Protocol: Model Context Protocol (MCP)