​
What are MCP Servers?

MCP Servers are the powerhouse behind agents in the mcp-agent framework. They provide specialized capabilities to agents through the Model Context Protocol (MCP), acting as external tools, data sources, and services that agents can access. Think of MCP servers as:
  • Tools that agents can call to perform specific tasks
  • Data sources that provide access to information and resources
  • Services that extend agent capabilities beyond the base LLM
  • Independent processes that can be developed, deployed, and scaled separately

Core Concept: MCP Servers extend agent capabilities by providing tools, resources, and prompts through a standardized protocol.

​
Server Types and Transports

The mcp-agent framework supports multiple transport mechanisms for connecting to MCP servers:

​
STDIO (Standard Input/Output)

Best for local development and subprocess-based servers: 
# mcp_agent.config.yaml
mcp:
  servers:
    filesystem:
      transport: "stdio" # Default transport
      command: "npx"
      args: ["-y", "@modelcontextprotocol/server-filesystem"]
      env:  # Environment variables passed to the server process
        ROOT_PATH: "/path/to/files"
      terminate_on_close: true # Default: true

​
Server-Sent Events (SSE)

Ideal for streaming responses and real-time data: 
mcp:
  servers:
    sse_server:
      transport: "sse"
      url: "http://localhost:8000/sse"
      headers:
        Authorization: "Bearer your-token"
      http_timeout_seconds: 30
      read_timeout_seconds: 60

​
WebSocket

For bidirectional, persistent connections: 
mcp:
  servers:
    websocket_server:
      transport: "websocket"
      url: "ws://localhost:8001/ws"
      headers:
        Authorization: "Bearer your-token"

​
Streamable HTTP

For HTTP-based servers with streaming support: 
mcp:
  servers:
    http_server:
      transport: "streamable_http"
      url: "http://localhost:8002/mcp"
      headers:
        Authorization: "Bearer your-token"
        Content-Type: "application/json"
      http_timeout_seconds: 30
      read_timeout_seconds: 120

​
Server Capabilities

MCP servers can provide three main types of capabilities:

​
1. Tools

Functions that agents can call to perform actions: 
# Example tool implementation using FastMCP
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("My Server")

@mcp.tool()
def calculate_sum(a: int, b: int) -> int:
    """Calculate the sum of two numbers."""
    return a + b

@mcp.tool()
def fetch_weather(city: str) -> str:
    """Get weather information for a city."""
    # Implementation here
    return f"Weather in {city}: Sunny, 75°F"

​
2. Resources

Data and content that agents can read and reference: 
@mcp.resource("file://{path}")
def read_file(path: str) -> str:
    """Read content from a file."""
    with open(path, 'r') as f:
        return f.read()

@mcp.resource("db://users/{user_id}")
def get_user(user_id: str) -> dict:
    """Get user information from database."""
    # Database lookup implementation
    return {"id": user_id, "name": "John Doe"}

​
3. Prompts

Reusable prompt templates that agents can utilize: 
@mcp.prompt()
def analysis_prompt(data: str, context: str = "") -> str:
    """Generate an analysis prompt with data and optional context."""
    return f"""Analyze the following data:

Data: {data}
Context: {context}

Provide a detailed analysis including key insights and recommendations."""

​
Configuration Examples

​
Basic Server Configuration

# mcp_agent.config.yaml
$schema: ../../schema/mcp-agent.config.schema.json

execution_engine: asyncio

mcp:
  servers:
    # Filesystem access
    filesystem:
      command: "npx"
      args: ["-y", "@modelcontextprotocol/server-filesystem"]
      env:  # Environment variables passed to the server process
        ROOT_PATH: "/workspace"

    # Web fetching capabilities
    fetch:
      command: "uvx"
      args: ["mcp-server-fetch"]

    # Custom SSE server
    analytics:
      url: "http://localhost:8000/sse"
      transport: "sse"
      headers:
        Authorization: "Bearer ${ANALYTICS_TOKEN}"

​
Advanced Server Configuration

mcp:
  servers:
    # Production database server with authentication
    database:
      name: "Production Database Server"
      description: "Provides access to production database"
      transport: "streamable_http"
      url: "https://api.example.com/mcp"
      headers:
        Authorization: "Bearer ${DB_API_TOKEN}"
        X-Client-ID: "${CLIENT_ID}"
      http_timeout_seconds: 30
      read_timeout_seconds: 120
      auth:
        type: "bearer"
        token: "${DB_API_TOKEN}"

    # Local development server with custom environment and roots
    dev_tools:
      name: "Development Tools Server"
      description: "Local development tools and utilities"
      transport: "stdio"
      command: "python"
      args: ["-m", "my_mcp_server"]
      env:  # Environment variables passed to the server process
        DEBUG: "true"
        LOG_LEVEL: "debug"
        DATABASE_URL: "${DEV_DATABASE_URL}"
      terminate_on_close: true
      roots:
        - uri: "file:///workspace"
          name: "Workspace"
        - uri: "file:///tmp"
          name: "Temporary Files"

​
Creating Your Own MCP Server

FastMCP provides the easiest way to create MCP servers: 
# server.py
from mcp.server.fastmcp import FastMCP
from mcp.server.models import InitializationOptions
import asyncio

# Create the server
mcp = FastMCP("My Custom Server")

@mcp.tool()
def greet(name: str) -> str:
    """Greet someone by name."""
    return f"Hello, {name}! Nice to meet you."

@mcp.tool()
async def async_calculation(x: float, y: float) -> float:
    """Perform an async calculation."""
    await asyncio.sleep(0.1)  # Simulate async work
    return x * y + 42

@mcp.resource("data://{dataset}")
def get_dataset(dataset: str) -> str:
    """Get dataset information."""
    datasets = {
        "sales": "Q1 Sales: $1.2M, Q2 Sales: $1.5M",
        "users": "Active Users: 15,432, New Users: 1,234"
    }
    return datasets.get(dataset, "Dataset not found")

@mcp.prompt()
def report_prompt(data_type: str, period: str = "monthly") -> str:
    """Generate a report prompt template."""
    return f"""Please create a {period} report for {data_type}.

Include:
1. Summary of key metrics
2. Trends and patterns
3. Recommendations for improvement
4. Action items for next period

Format the report in a clear, professional manner."""

# Run the server
if __name__ == "__main__":
    mcp.run()

​
Integration Patterns

​
Using Servers with Agents

from mcp_agent.agents.agent import Agent
from mcp_agent.workflows.llm.augmented_llm_openai import OpenAIAugmentedLLM

# Create an agent with multiple server types
agent = Agent(
    name="data_analyst",
    instruction="""You are a data analyst with access to databases,
    file systems, and analytics tools. Help users analyze data and
    generate insights.""",
    server_names=["filesystem", "database", "analytics"]
)

async with agent:
    # Discover available tools
    tools = await agent.list_tools()
    print(f"Available tools: {[tool.name for tool in tools.tools]}")

    # Use the agent with an LLM
    llm = await agent.attach_llm(OpenAIAugmentedLLM)

    result = await llm.generate_str(
        "Analyze the sales data from Q1 and create a summary report"
    )
    print(result)

​
Server Development Best Practices

​
1. Error Handling

@mcp.tool()
def safe_division(a: float, b: float) -> str:
    """Safely divide two numbers."""
    try:
        if b == 0:
            return "Error: Division by zero is not allowed"
        result = a / b
        return f"Result: {result}"
    except Exception as e:
        return f"Error: {str(e)}"

​
2. Input Validation

from pydantic import BaseModel, validator

class WeatherRequest(BaseModel):
    city: str
    units: str = "fahrenheit"

    @validator('city')
    def city_must_not_be_empty(cls, v):
        if not v.strip():
            raise ValueError('City name cannot be empty')
        return v.strip()

    @validator('units')
    def units_must_be_valid(cls, v):
        if v not in ['fahrenheit', 'celsius']:
            raise ValueError('Units must be fahrenheit or celsius')
        return v

@mcp.tool()
def get_weather(request: WeatherRequest) -> str:
    """Get weather with validated input."""
    # Implementation here
    return f"Weather in {request.city}: 75°{request.units[0].upper()}"

​
3. Async Operations

import aiohttp

@mcp.tool()
async def fetch_url(url: str) -> str:
    """Fetch content from a URL asynchronously."""
    async with aiohttp.ClientSession() as session:
        try:
            async with session.get(url) as response:
                if response.status == 200:
                    content = await response.text()
                    return f"Content fetched successfully (length: {len(content)})"
                else:
                    return f"Error: HTTP {response.status}"
        except Exception as e:
            return f"Error fetching URL: {str(e)}"

​
Advanced Features

​
Elicitation Support

Elicitation allows servers to request additional structured input from users during tool execution: 
from mcp.server.fastmcp import FastMCP, Context
from mcp.server.elicitation import (
    AcceptedElicitation,
    DeclinedElicitation,
    CancelledElicitation,
)
from pydantic import BaseModel, Field

mcp = FastMCP("Booking System")

@mcp.tool()
async def book_table(date: str, party_size: int, ctx: Context) -> str:
    """Book a table with confirmation"""
    
    # Schema must only contain primitive types (str, int, float, bool)
    class ConfirmBooking(BaseModel):
        confirm: bool = Field(description="Confirm booking?")
        notes: str = Field(default="", description="Special requests")
    
    result = await ctx.elicit(
        message=f"Confirm booking for {party_size} on {date}?", 
        schema=ConfirmBooking
    )
    
    match result:
        case AcceptedElicitation(data=data):
            if data.confirm:
                return f"Booked! Notes: {data.notes or 'None'}"
            return "Booking cancelled"
        case DeclinedElicitation():
            return "Booking declined"
        case CancelledElicitation():
            return "Booking cancelled"

​
Production Considerations

​
Security

# Use environment variables for sensitive data
import os

@mcp.tool()
def secure_api_call(endpoint: str) -> str:
    """Make a secure API call using stored credentials."""
    api_key = os.getenv("API_KEY")
    if not api_key:
        return "Error: API key not configured"

    # Make authenticated request
    # Implementation here
    return "API call completed successfully"

​
Performance

# Configure timeouts and headers for better performance
mcp:
  servers:
    high_traffic_server:
      transport: "streamable_http"
      url: "https://api.example.com/mcp"
      http_timeout_seconds: 30
      read_timeout_seconds: 120
      headers:
        Keep-Alive: "timeout=60, max=100"
        Connection: "keep-alive"

​
Monitoring

import logging

# Configure logging in your server
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

@mcp.tool()
def monitored_operation(data: str) -> str:
    """Operation with monitoring and logging."""
    logger.info(f"Starting operation with data length: {len(data)}")

    try:
        # Process data
        result = process_data(data)
        logger.info("Operation completed successfully")
        return result
    except Exception as e:
        logger.error(f"Operation failed: {str(e)}")
        return f"Error: {str(e)}"

Getting Started

Explore example MCP servers and learn implementation patterns.

FastMCP Documentation

Learn more about FastMCP and the official MCP server toolkit.

Agent Integration

Learn how agents discover and use MCP server capabilities.