fastapi_mcp_template

fastapi_mcp_template

3.2

If you are the rightful owner of fastapi_mcp_template and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

A production-ready FastAPI MCP server template for rapid MCP development and prototyping.

FastAPI MCP Template

A production-ready FastAPI MCP server template for rapid MCP development and prototyping

This template provides a complete foundation for building custom MCP servers with dynamic tool mounting, dual-interface support (MCP + REST API), and containerized deployment. The main goal is to deliver a dynamically flexible template for fast MCP development - allowing developers to quickly create powerful MCP servers by simply adding tools to the tools/ directory.

Featured Example: Advanced File-to-Markdown Conversion

This template demonstrates its capabilities with sophisticated file conversion tools that include:

  • Standard file conversion using MarkItDown library
  • LLM-enhanced image processing for detailed descriptions
  • Multi-level OCR fallback with traditional OCR → LLM vision analysis
  • Smart processing optimization balancing cost and accuracy
  • Support for multiple file formats: PDF, DOCX, images, and more

🎯 Project Goal

Enable rapid MCP server development through a flexible, self-contained template architecture.

Instead of building MCP servers from scratch, developers can:

  1. Clone this template
  2. Add their custom tools to the tools/ directory
  3. Deploy immediately with Docker
  4. Focus on business logic rather than infrastructure

As a demonstration, I've included advanced file conversion tools as an example, but this template can be adapted for any MCP use case: data processing, API integrations, content generation, analysis tools, and more.

🏗️ Architecture Overview

fastapi-mcp-template/
├── fastapi_mcp_template/          # Core MCP server package
│   ├── __about__.py              # Version information
│   ├── __init__.py               # Package initialization
│   ├── config.py                 # Configuration management
│   ├── main.py                   # Application entry point
│   ├── requirements.txt          # Core dependencies only
│   ├── api/                      # REST API endpoints
│   │   ├── __init__.py
│   │   └── routes.py             # API route definitions
│   ├── core/                     # MCP core functionality
│   │   ├── __init__.py
│   │   ├── tool_base.py          # Base tool infrastructure
│   │   ├── tool_definition.py    # Tool interface definitions
│   │   └── tool_manager.py       # Dynamic tool loading
│   └── core_test/                # Core functionality tests
│       ├── __init__.py
│       ├── requirements.txt      # Test dependencies
│       ├── test_base.py
│       ├── test_core.py
│       └── test_manager.py
├── tools/                        # Self-contained tools (your custom tools go here!)
│   ├── file_converter.py         # Example: File-to-Markdown converter with OCR
│   ├── text_processor.py         # Example: Text processing operations
│   ├── url_fetcher.py            # Example: URL content fetcher
│   └── requirements/             # Tool-specific dependencies
│       ├── file_converter.txt    # Dependencies for file_converter (markitdown, openai, PyMuPDF)
│       ├── text_processor.txt    # Dependencies for text_processor
│       └── url_fetcher.txt       # Dependencies for url_fetcher
├── tests/                        # Tool tests
│   ├── __init__.py
│   └── test_tools.py
├── data/                         # Data directory for file storage
├── docker-compose.yml            # Container orchestration
├── Dockerfile                    # Container definition with dynamic deps
├── .env.example                  # Environment configuration template
└── README.md                     # This file

🚀 Key Features

Dynamic Tool Loading

  • Zero Configuration: Drop Python files in tools/ and they're automatically discovered
  • Hot Reloading: Development mode automatically reloads tools when changed
  • Dependency Isolation: Each tool manages its own dependencies via requirements/{toolname}.txt
  • Self-Contained: Tools have no dependencies on the core package
  • Dynamic Tool Types: Tools can define any custom type without core modifications

Dual Interface Architecture

  • MCP Protocol: Native integration with LLM applications (Claude, etc.)
  • REST API: Standard HTTP endpoints for web applications
  • Unified Tool Access: Same tools available through both interfaces
  • Interactive Documentation: Auto-generated API docs at /docs

Production-Ready Deployment

  • Docker Support: Complete containerization with multi-stage builds
  • Dynamic Dependencies: Tool dependencies installed only when tools are enabled
  • Health Checks: Built-in monitoring and health endpoints
  • Environment Configuration: Flexible configuration through environment variables

Developer Experience

  • FastAPI Foundation: Modern Python web framework with automatic validation
  • Type Safety: Full type hints and Pydantic models
  • Error Handling: Comprehensive error handling and logging
  • Testing Ready: Structure supports easy unit and integration testing

📋 MCP Protocol Compliance

This template implements the Model Context Protocol (MCP) 2024-11-05 specification with full compliance to ensure seamless integration with MCP clients like Claude Desktop, Cline, and other LLM applications.

Protocol Standards Adherence

JSON-RPC 2.0 Foundation
  • ✅ Full JSON-RPC 2.0 compliance with proper message structure
  • ✅ Unique request IDs that are never reused within a session
  • ✅ Proper error codes following JSON-RPC standards (-32000 to -32099 range)
  • ✅ Structured error responses with code, message, and optional data fields
MCP Lifecycle Management
  • ✅ Proper initialization handshake with capability negotiation
  • ✅ Session management with UUID-based session IDs
  • ✅ Protocol version negotiation (supports 2024-11-05)
  • ✅ Graceful error handling for unsupported protocol versions
Core MCP Methods
{
  "initialize": "✅ Capability negotiation with tool listing support",
  "notifications/initialized": "✅ Post-initialization state management", 
  "tools/list": "✅ Dynamic tool discovery and enumeration",
  "tools/call": "✅ Tool execution with parameter validation",
  "ping": "✅ Connection liveness verification"
}

Capability Declaration

The server correctly declares its capabilities during initialization:

{
  "capabilities": {
    "tools": {
      "listChanged": true
    }
  },  
  "serverInfo": {
    "name": "fastapi_mcp_template",
    "version": "1.0.0"
  }
}

Key Compliance Features:

  • Tool Capability: Declares support for tool listing and execution
  • Dynamic Updates: Supports listChanged notifications for tool modifications
  • Server Identity: Proper server information with name and version
  • Protocol Version: Explicit protocol version support declaration

Tool Schema Compliance

Tools follow MCP tool specification with proper JSON Schema validation:

{
  "name": "file_to_markdown",
  "description": "Convert files to Markdown format",
  "inputSchema": {
    "type": "object",
    "properties": {
      "filename": {
        "type": "string",
        "description": "Name of the file to convert"
      },
      "base64_content": {
        "type": "string", 
        "description": "Base64 encoded file content"
      }
    },
    "required": ["filename", "base64_content"]
  }
}

Error Handling Standards

Protocol-Level Errors
  • -32601: Method not found (unsupported MCP methods)
  • -32602: Invalid params (malformed requests, unknown tools)
  • -32603: Internal error (server-side failures)
  • -32002: Server not initialized (session/capability issues)
Tool Execution Errors

Tool errors are reported within successful JSON-RPC responses using the isError flag:

{
  "jsonrpc": "2.0",
  "id": 123,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Tool execution failed: File format not supported"
      }
    ],
    "isError": true
  }
}

Session Management

Session Lifecycle
  1. Initialization: Client sends initialize request
  2. Capability Negotiation: Server responds with supported features
  3. Ready State: Client sends notifications/initialized
  4. Active Session: Tools can be listed and called
  5. Session Persistence: Sessions maintain state across requests
Session Security
  • UUID-based Session IDs: Cryptographically secure session identifiers
  • Session Validation: All requests validated against active sessions
  • Session Timeout: Automatic cleanup of inactive sessions
  • Header-based Authentication: Session ID passed via Mcp-Session-Id header

Transport Layer Compliance

HTTP Transport Support
  • ✅ Content-Type: application/json for all requests/responses
  • ✅ CORS Support: Proper cross-origin headers for web clients
  • ✅ Session Headers: Mcp-Session-Id header handling
  • ✅ Status Codes: Appropriate HTTP status codes (200, 202, 400, 500)
Endpoint Structure
POST /mcp                    # Main MCP protocol endpoint
POST /                      # Fallback MCP endpoint
GET  /mcp/session           # Session creation helper
OPTIONS /mcp                # CORS preflight support

Testing & Validation

MCP Inspector Compatibility

The server is fully compatible with the official MCP Inspector tool:

# Test with MCP Inspector
npx @modelcontextprotocol/inspector
# Connect to: http://localhost:8000/mcp
Protocol Validation Tests
  • ✅ Initialization Flow: Complete handshake sequence
  • ✅ Tool Discovery: List and validate all available tools
  • ✅ Tool Execution: Parameter validation and result formatting
  • ✅ Error Scenarios: Proper error code and message handling
  • ✅ Session Management: Session creation, validation, and cleanup

Interoperability

Client Compatibility

Tested and verified with:

  • ✅ Claude Desktop: Full integration support
  • ✅ MCP Inspector: Official testing tool compatibility
  • ✅ Custom Clients: Java, TypeScript, Python client libraries
  • ✅ Web Clients: Browser-based MCP applications
Standards Compliance
  • 📋 MCP Specification 2024-11-05: Full compliance
  • 📋 JSON-RPC 2.0: Complete implementation
  • 📋 JSON Schema: Tool parameter validation
  • 📋 HTTP/1.1: Standard web compatibility
  • 📋 CORS: Cross-origin resource sharing support

This comprehensive compliance ensures that any MCP client can seamlessly connect to and interact with your server, providing a reliable foundation for LLM integrations.

🛠️ Example Use Cases

This template can be adapted for any MCP server use case:

  • 📄 Document Processing: Convert, analyze, or transform documents with LLM-enhanced OCR
  • 🔗 API Integrations: Connect to external services and APIs
  • 📊 Data Analysis: Process and analyze datasets
  • 🎨 Content Generation: Create or transform content with AI assistance
  • 🔍 Information Retrieval: Search and fetch information from various sources
  • 🧮 Calculations: Perform complex calculations or simulations
  • 📈 Monitoring: Gather metrics or monitor systems
  • 🖼️ Image Analysis: Extract text and descriptions from images using vision-capable LLMs

📦 Installation & Setup

Option 1: Docker (Recommended)

Quick Start - All tools enabled:

# Clone the template
git clone https://github.com/SteffenHebestreit/fastapi-mcp-template.git
cd fastapi-mcp-template

# Start with default configuration
docker-compose up --build

Custom tool selection:

# Copy environment template
cp .env.example .env

# Edit .env to specify which tools to enable
# ENABLED_TOOLS=file_converter,text_processor

# Build with custom tools
docker-compose up --build

Advanced Docker usage:

# Build with specific tools only
docker-compose build --build-arg ENABLED_TOOLS=file_converter,text_processor

# Run in background
docker-compose up -d

# View logs
docker-compose logs -f

# Stop services
docker-compose down

Option 2: Direct Installation

For development or custom deployments:

# 1. Clone and navigate
git clone https://github.com/SteffenHebestreit/fastapi-mcp-template.git
cd fastapi-mcp-template

# 2. Install core dependencies
pip install -r fastapi_mcp_template/requirements.txt

# 3. Install tool dependencies (choose what you need)
pip install -r tools/requirements/file_converter.txt  # For file conversion with enhanced OCR
pip install -r tools/requirements/url_fetcher.txt     # For URL fetching
# text_processor has no additional dependencies

# 4. Additional OCR setup (for traditional OCR fallback)
# For pytesseract: Install Tesseract OCR engine
# - Windows: Download from https://github.com/UB-Mannheim/tesseract/wiki
# - Ubuntu/Debian: sudo apt install tesseract-ocr
# - macOS: brew install tesseract

# 5. Run the server
python fastapi_mcp_template/main.py

# Or with custom configuration
python fastapi_mcp_template/main.py --host 0.0.0.0 --port 8000

🔧 Configuration

Environment Variables

Configure the server behavior through environment variables:

# Core Configuration
HOST=0.0.0.0                    # Server host
PORT=8000                       # Server port
DEBUG=false                     # Debug mode
LOG_LEVEL=INFO                  # Logging level

# Tool Configuration  
ENABLED_TOOLS=file_converter,text_processor,url_fetcher  # Comma-separated list
TOOLS_DIR=/app/tools            # Tools directory path
MAX_FILE_SIZE=10485760          # Max file size (10MB)

# API Configuration
API_TITLE="FastAPI MCP Template"
API_VERSION="1.0.0"
DOCS_URL="/docs"                # API documentation URL

OpenAI Integration Configuration

Enable OpenAI integration for enhanced image processing and content analysis:

Basic OpenAI Setup
# Enable LLM features for enhanced file conversion
FTMD_MARKITDOWN_ENABLE_LLM=true

# Your OpenAI API key
FTMD_OPENAI_API_KEY=sk-proj-your-api-key-here

# Model to use (default: gpt-4o)
FTMD_OPENAI_MODEL=gpt-4o
Azure OpenAI Setup
# Enable LLM features
FTMD_MARKITDOWN_ENABLE_LLM=true

# Azure OpenAI endpoint
FTMD_OPENAI_BASE_URL=https://your-resource.openai.azure.com/

# Azure API key
FTMD_OPENAI_API_KEY=your-azure-api-key

# Model deployment name
FTMD_OPENAI_MODEL=gpt-4o
Custom OpenAI Compatible Endpoints
# Enable LLM features
FTMD_MARKITDOWN_ENABLE_LLM=true

# Custom endpoint
FTMD_OPENAI_BASE_URL=https://your-custom-endpoint.com/v1

# API key for custom endpoint
FTMD_OPENAI_API_KEY=your-custom-api-key

# Model name
FTMD_OPENAI_MODEL=your-model-name
Supported Models

Common OpenAI models that work well with MarkItDown:

  • gpt-4o (recommended)
  • gpt-4o-mini (cost-effective)
  • gpt-4-turbo
  • gpt-4
Features Enhanced by LLM Integration

When LLM integration is enabled, the file converter provides:

  1. Enhanced Image Processing: Generate detailed descriptions of images, charts, and diagrams
  2. OCR for Scanned PDFs: Automatically converts PDF pages to images and extracts text using vision-capable LLMs
  3. Better Content Analysis: Improved understanding of document structure and content
  4. Smart Formatting: More intelligent Markdown formatting decisions

Enhanced OCR Fallback Process: The file converter now uses a sophisticated three-level fallback system for scanned PDFs:

  1. Traditional OCR (Primary): Fast and cost-effective text extraction using pytesseract or easyocr

    • Processes document images using traditional OCR algorithms
    • Ideal for clear, standard fonts and simple layouts
    • No API costs, works offline

  2. LLM Vision Analysis (Secondary): Advanced analysis for complex content

    • Used when traditional OCR yields insufficient results (<50 characters)
    • Handles handwriting, complex layouts, and poor-quality scans
    • Uses vision-capable LLMs for superior accuracy

  3. Intelligent Selection (Tertiary): Combines best results

    • Automatically selects the most comprehensive output
    • Prefers traditional OCR when sufficient, LLM when needed
    • Optimizes for both cost and accuracy

Process Details:

  • Automatically detects scanned PDFs (empty text content)
  • Converts PDF pages to high-resolution images (2x zoom)
  • Processes up to 5 pages per document to manage costs
  • Maintains original formatting and structure in Markdown output

Docker Configuration

docker-compose.yml supports:

  • Tool selection via build args
  • Environment file mounting
  • Volume mounting for development
  • Port configuration
  • Health checks

Dockerfile features:

  • Multi-stage builds for optimization
  • Dynamic dependency installation based on enabled tools
  • Security-hardened container (non-root user)
  • Health check endpoints

🎯 Adding Your Own Tools

Dynamic Tool Type System

The template uses a flexible, dynamic tool type system that allows you to define any custom tool type without modifying core code.

How It Works

Tools use simple strings for types instead of hardcoded enums:

# Define any tool type you want
tool_type = "converter"     # Built-in type
tool_type = "analyzer"      # Custom type
tool_type = "my_special_ai" # Your unique type
Available Built-in Types

For convenience, common tool types are available:

  • "converter" - File format converters
  • "processor" - Text/data processors
  • "fetcher" - Content/data fetchers
  • "utility" - General utility tools
Custom Types

You can create any custom type:

def get_definition(self):
    return ToolDefinition(
        name="sentiment_analyzer",
        description="AI-powered sentiment analysis",
        endpoint="/sentiment",
        tool_type="ai_analyzer",  # Custom type!
        version="1.0.0"
    )

The system automatically discovers and tracks all tool types as tools are loaded.

Step 1: Create a Tool File

Create a new Python file in the tools/ directory:

# tools/my_custom_tool.py
from typing import Dict, Any

class ToolInterface:
    """Interface that tools must implement."""
    def get_definition(self): raise NotImplementedError
    def get_schema(self): raise NotImplementedError
    async def execute(self, **kwargs) -> Dict[str, Any]: raise NotImplementedError

class ToolDefinition:
    def __init__(self, name: str, description: str, endpoint: str, tool_type: str, version: str = "1.0.0"):
        self.name = name
        self.description = description
        self.endpoint = endpoint
        self.tool_type = tool_type
        self.version = version

class ToolSchema:
    def __init__(self, properties: Dict[str, Any], required: list = None):
        self.properties = properties
        self.required = required or []

class MyCustomTool(ToolInterface):
    def __init__(self, tool_base):
        self.tool_base = tool_base
    
    def get_definition(self) -> ToolDefinition:
        return ToolDefinition(
            name="my_custom_tool",
            description="Description of what my tool does",
            endpoint="/my-endpoint",
            tool_type="my_analyzer",  # Any string you want!
            version="1.0.0"
        )
    
    def get_schema(self) -> ToolSchema:
        return ToolSchema(
            properties={
                "input_param": {
                    "type": "string",
                    "description": "Description of the parameter"
                }
            },
            required=["input_param"]
        )
    
    async def execute(self, **kwargs) -> Dict[str, Any]:
        input_param = kwargs.get("input_param")
        
        try:
            # Your tool logic here
            result = f"Processed: {input_param}"
            
            self.tool_base.log_info(f"Tool executed successfully")
            return {
                "success": True,
                "result": result
            }
        except Exception as e:
            self.tool_base.log_error(f"Tool execution failed: {str(e)}")
            return {
                "success": False,
                "error": str(e)
            }

def setup_tool(tool_base):
    """Setup function called by the core service."""
    return MyCustomTool(tool_base)

Step 2: Add Dependencies (if needed)

If your tool requires additional packages:

# Create tools/requirements/my_custom_tool.txt
mkdir -p tools/requirements
echo "requests>=2.28.0" > tools/requirements/my_custom_tool.txt
echo "beautifulsoup4>=4.11.0" >> tools/requirements/my_custom_tool.txt

Step 3: Enable Your Tool

Add your tool to the enabled tools list:

# In .env file or environment
ENABLED_TOOLS=file_converter,text_processor,url_fetcher,my_custom_tool

Step 4: Deploy

# Rebuild and restart
docker-compose up --build

That's it! Your tool is now available through both MCP and REST API interfaces.

🌐 API Endpoints

Once running, the server provides multiple interfaces:

MCP Protocol Endpoints

  • POST /mcp - Main MCP protocol endpoint (handles all MCP methods)
  • GET /mcp/session - Session creation helper endpoint

REST API Endpoints

  • GET /api/health - Health check endpoint
  • GET /api/tools - List all available tools
  • POST /api/tools/{tool_name} - Execute a specific tool
  • GET /api/tools/{tool_name}/schema - Get tool parameter schema
  • GET /docs - Interactive API documentation

Available Tools

  • file_to_markdown - Convert files to Markdown with OCR fallback
  • text_processor - Process and transform text content
  • url_fetcher - Fetch and process content from URLs

Example Usage

Via REST API:

# List available tools
curl http://localhost:8000/api/tools

# Execute file converter with base64 content
curl -X POST http://localhost:8000/api/tools/file_to_markdown \
  -H "Content-Type: application/json" \
  -d '{"filename": "document.pdf", "base64_content": "JVBERi0xLjQK..."}'

# Process text
curl -X POST http://localhost:8000/api/tools/text_processor \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello World", "operation": "uppercase"}'

# Fetch URL content
curl -X POST http://localhost:8000/api/tools/url_fetcher \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

Via MCP Protocol:

# Initialize MCP session
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {"name": "test-client", "version": "1.0.0"}
    }
  }'

# List tools (after initialization)
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Mcp-Session-Id: <session-id-from-initialize>" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/list",
    "params": {}
  }'

# Call tool
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Mcp-Session-Id: <session-id>" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "file_to_markdown",
      "arguments": {"filename": "test.pdf", "base64_content": "..."}
    }
  }'

🔍 Monitoring & Health

  • Health Check: GET /api/health
  • Metrics: Built-in logging and metrics collection
  • Docker Health: Container health checks configured
  • API Documentation: Interactive docs at /docs

🧪 Development & Testing

Development Mode

# Run with hot reloading
python fastapi_mcp_template/main.py --reload

# Or with Docker in development mode
docker-compose -f docker-compose.yml -f docker-compose.dev.yml up

Testing the MCP Server

Using MCP Inspector (Recommended)

  1. Make sure the fastapi-mcp-template server is running:

    docker-compose up -d

  2. Run the MCP Inspector to test the server:

    npx @modelcontextprotocol/inspector

  3. In the MCP Inspector UI:

Manual Testing with curl

Test the initialization endpoint:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {
        "name": "test-client",
        "version": "1.0.0"
      }
    }
  }'

Test the tools/list endpoint (after initialization):

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Mcp-Session-Id: <session-id-from-initialize>" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/list",
    "params": {}
  }'
Expected Behavior
  1. Initialize should return a session ID and server info
  2. The initialized notification should be sent automatically
  3. Tools/list should return available tools after proper initialization
  4. The server should not return "Server not initialized" errors
Debugging
  • Check Docker logs: docker-compose logs fastapi-mcp-template
  • Check if port 8000 is accessible: curl http://localhost:8000/mcp
  • Use MCP Inspector's debug features to see the full request/response flow
Testing LLM-Enhanced File Conversion

Test the advanced file conversion capabilities:

Convert an image with enhanced descriptions:

curl -X POST "http://localhost:8000/api/tools/file_to_markdown" \
  -H "Content-Type: application/json" \
  -d '{
    "filename": "chart.png",
    "base64_content": "iVBORw0KGgoAAAANSUhEUgAA..."
  }'

Convert a scanned PDF with OCR fallback:

curl -X POST "http://localhost:8000/api/tools/file_to_markdown" \
  -H "Content-Type: application/json" \
  -d '{
    "filename": "scanned_document.pdf", 
    "base64_content": "JVBERi0xLjYNJeLjz9MN..."
  }'

Expected behavior with LLM integration:

  • Images: Detailed descriptions of visual content
  • Scanned PDFs: Automatic OCR processing when no text is found
  • Regular PDFs: Standard text extraction with enhanced formatting

LLM Integration Troubleshooting:

  • PyMuPDF Not Available: Install with pip install PyMuPDF>=1.24.0 for OCR functionality
  • OpenAI Library Not Available: Install dependencies with pip install openai>=1.0.0
  • API Key Issues:
    • Ensure your API key is valid and has sufficient credits
    • Check that the base URL is correct for Azure OpenAI or custom endpoints
    • Verify environment variable names use the FTMD_ prefix
  • Model Not Found:
    • Ensure the model name is correct for your OpenAI setup
    • For Azure OpenAI, use your deployment name, not the base model name
    • Verify the model supports vision capabilities for OCR features
  • OCR Not Working:
    • Check that FTMD_MARKITDOWN_ENABLE_LLM=true is set
    • Verify the PDF actually contains no extractable text (scanned documents)
    • Look for "Attempting enhanced OCR fallback" messages in logs
  • Traditional OCR Issues:
    • Pytesseract: Ensure Tesseract OCR engine is installed on system
    • EasyOCR: May require additional GPU setup for optimal performance
    • Low Quality Results: Tool will automatically fall back to LLM vision analysis
  • OCR Performance:
    • Traditional OCR is preferred for cost efficiency (no API costs)
    • LLM vision analysis activates when OCR yields <50 characters
    • Processing limited to 5 pages per PDF to manage costs

Security Considerations:

  • Never commit API keys to version control
  • Use environment variables or secure secret management
  • Rotate API keys regularly
  • Monitor API usage for unexpected activity

Cost Considerations:

  • LLM integration incurs API costs based on usage
  • OCR processing (scanned PDFs) uses more tokens than standard text extraction
  • Image processing typically uses more tokens than text-only operations
  • Consider using gpt-4o-mini for cost-effective processing
  • OCR processing is limited to 5 pages per PDF to manage costs
  • Monitor your usage through the OpenAI dashboard

Testing Structure

The project uses a separated testing approach that mirrors the separation between core functionality and tools.

Test Organization

1. Core Tests (fastapi_mcp_template/core_test/)

  • Purpose: Tests for the core MCP Template Server functionality
  • Location: Inside the template service package
  • Covers: Tool manager, configuration loading, API routes, core application logic

2. Tool Tests (tests/)

  • Purpose: Tests specifically for individual tools
  • Location: Root-level tests directory
  • Covers: Individual tool loading and execution, tool-specific functionality, tool requirements validation
Running Tests

Via Docker Compose (Development):

# Start development container with tests enabled
docker-compose --profile dev up fastapi-mcp-template-dev

# List available tests
curl http://localhost:8000/tests

# Check test status
curl http://localhost:8000/tests/status

# Run all tests via API
curl -X POST http://localhost:8000/tests/run-all

# Run specific test
curl -X POST http://localhost:8000/tests/run/test_tools

Via Environment Variable:

# Enable tests in environment
export ENABLE_TESTS=true

# Or in .env file
echo "ENABLE_TESTS=true" >> .env

Direct Test Execution:

# Run core tests
cd fastapi_mcp_template
python -m unittest discover core_test

# Run tool tests  
cd tests
python -m unittest discover .
Test API Endpoints

When tests are enabled (ENABLE_TESTS=true), the following endpoints are available:

  • GET /tests/status - Check if testing is enabled
  • GET /tests - List all available tests
  • POST /tests/run/{test_name} - Run specific test file
  • POST /tests/run-all - Run all tests
Adding New Tests

For Core Functionality: Add test files to fastapi_mcp_template/core_test/ following the pattern test_*.py

For Tools: Add test files to tests/ following the pattern test_*.py

Both types use the same base testing classes for consistency.

Tool Development Tips

  1. Keep Tools Self-Contained: No imports from the core package
  2. Handle Errors Gracefully: Always return structured error responses
  3. Use Type Hints: Leverage Python's type system for better development experience
  4. Log Operations: Use tool_base.log_info() and tool_base.log_error()
  5. Validate Inputs: Implement proper input validation in your tools

📝 License

This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License.

What this means:

  • ✅ Attribution Required: You must give appropriate credit to the original creator
  • ✅ Non-Commercial Use: You can use, modify, and distribute for non-commercial purposes
  • ✅ Remix & Adapt: You can build upon and modify the material
  • ❌ Commercial Use: Commercial use requires separate licensing agreement

Copyright (c) 2025 Steffen Hebestreit

For the full license text visit: https://creativecommons.org/licenses/by-nc/4.0/

For commercial licensing, please contact the author.

🚀 Getting Started

Ready to build your MCP server?

  1. Fork this template
  2. Clone your fork 
    git clone https://github.com/yourusername/fastapi-mcp-template.git
cd fastapi-mcp-template
  3. Add your custom tools to the tools/ directory
  4. Configure via .env file
  5. Deploy with docker-compose up --build

Your MCP server will be ready to integrate with LLM applications and provide both MCP and REST API access to your tools!

Happy building! 🛠️