FEL MCP Server (Local)

Local Model Context Protocol (MCP) server for Guatemalan FEL invoices.
Exposes tools to validate FEL XML and render branded PDFs. Designed for WSL (Ubuntu) or Linux and can be consumed by MCP hosts (e.g., Claude Desktop).

This repository focuses on the FEL MCP server. It’s often used alongside a chatbot client; see “Related repositories”.

🔗 Related repositories

• Chatbot (CLI / UI) — client that can drive this MCP server.

• Reference: OpenAI Chat API Example — reference only (patterns for API connectivity and instruction context).

Related MCP server: Dokmatiq DocGen

✨ Tools (MCP)

• fel_validate
  required: xml_path (absolute WSL path)
  Checks required fields and totals (subtotal, VAT 12%, total).

• fel_render
  required: xml_path, out_path (absolute WSL paths)
  Renders a branded PDF from the FEL XML.

• fel_batch
  required: dir_xml, out_dir (absolute WSL paths)
  Processes all *.xml in dir_xml, produces one PDF per XML in out_dir, and writes manifest.json.

Use absolute WSL paths like /mnt/d/... in all arguments and environment variables.

⚙️ Requirements

• Python 3.12

• WSL Ubuntu 22.04 (or Linux)

• Virtual environment (recommended)

🔧 Installation

git clone <REPO_URL>
cd <REPO_DIR>

python3.12 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Create your environment file (use absolute WSL paths):

cp .env.example .env

Key variables used by the server (placeholders shown):

# I/O
FEL_XML_PATH=/ABSOLUTE/PATH/TO/REPO/data/xml/factura.xml
FEL_OUTPUT_PDF=/ABSOLUTE/PATH/TO/REPO/data/out/factura.pdf
FEL_BATCH_OUT_DIR=/ABSOLUTE/PATH/TO/REPO/data/out
# optional logo
FEL_LOGO_PATH=/ABSOLUTE/PATH/TO/REPO/data/logos/logo.jpg

# Fonts
FEL_ACTIVE_FONT=1
FEL_FONT_DIR_MONTSERRAT=/ABSOLUTE/PATH/TO/REPO/assets/fonts/Montserrat/static
FEL_FONT_DIR_ROBOTOMONO=/ABSOLUTE/PATH/TO/REPO/assets/fonts/Roboto_Mono/static

# Theme/Layout
FEL_THEME=light
FEL_QR_SIZE=150
FEL_TOP_BAR_HEIGHT=20

▶️ Run the server (STDIO)

source venv/bin/activate
python servers/fel_mcp_server/server_stdio.py

🧪 Quick CLI tests (JSON-RPC over stdin)

List tools:

printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/list"}' \
| python servers/fel_mcp_server/server_stdio.py

Validate one XML:

printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"fel_validate","arguments":{"xml_path":"/ABSOLUTE/PATH/TO/REPO/data/xml/factura.xml"}}}' \
| python servers/fel_mcp_server/server_stdio.py

Render one PDF (no logo):

printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"fel_render","arguments":{"xml_path":"/ABSOLUTE/PATH/TO/REPO/data/xml/factura.xml","out_path":"/ABSOLUTE/PATH/TO/REPO/data/out/testing.pdf"}}}' \
| python servers/fel_mcp_server/server_stdio.py

Batch:

printf '%s\n' \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' \
'{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"fel_batch","arguments":{"dir_xml":"/ABSOLUTE/PATH/TO/REPO/data/xml","out_dir":"/ABSOLUTE/PATH/TO/REPO/data/out/batch"}}}' \
| python servers/fel_mcp_server/server_stdio.py

🖥️ Use with Claude Desktop (MCP)

1. Install Claude Desktop: https://claude.ai/download

2. Open Settings -> Developer -> Edit Config, then edit claude_desktop_config.json:

   {
     "mcpServers": {
       "FEL": {
         "command": "wsl.exe",
         "args": [
           "-e",
           "/ABSOLUTE/PATH/TO/REPO/venv/bin/python",
           "/ABSOLUTE/PATH/TO/REPO/servers/fel_mcp_server/server_stdio.py"
         ]
       }
     }
   }

   • Use absolute WSL paths in args.

   • If the host logs show errors like "'xml_path'", it means the call was sent without arguments; re-run with a prompt that includes the exact JSON arguments block.

3. Restart Claude Desktop (PowerShell):

Stop-Process -Name "Claude" -Force; Start-Process "<ABSOLUTE_WINDOWS_PATH_TO_Claude.exe>"

📚 References

• Model Context Protocol

• Anthropic API

• JSON-RPC 2.0