[!][ ANNOUNCEMENT ]|Introducing AgentML—A deterministic language for agent behaviorLEARN MORE

SYSTEM_STATUS: PREPARING_LAUNCH

Build and deploy reliable PRODUCTION AGENTS

System_Info

We turn agent experiments into production-grade systems with versioned agents, permissioned subagents, and full-stack observability so teams can safely run AI workflows in production.

We're building the foundation for agents that don't just think. They work.

Introducing AgentML—A deterministic language for agent behavior. MIT licensed, open source, and maintained by Agentflare.Learn More

Back to Blog

January 21, 2025•Justin Merrell

MCP: The Universal Protocol for AI Tool Servers

A deep dive into the Model Context Protocol—how it works, why it matters, and how to implement it for your AI agents.

mcptool-serversai-agentsprotocolstutorialbest-practices

The USB-C Moment for AI

Remember the chaos of proprietary phone chargers? Every device needed its own cable, creating a drawer full of incompatible connectors. The AI agent ecosystem faces the same fragmentation today—except instead of charging cables, it's tool integration protocols.

Enter the Model Context Protocol (MCP): Anthropic's answer to the AI integration chaos. Like USB-C revolutionized device connectivity, MCP promises to standardize how AI agents connect to tools, databases, and external systems. No more building custom integrations for every AI provider. No more maintaining multiple protocol implementations. Just one standard that works everywhere.

But MCP is more than just another protocol—it's becoming the foundation for the next generation of AI agent infrastructure. With adoption by OpenAI, Google, and Microsoft in 2025, MCP is rapidly becoming the industry standard. Here's everything you need to know to implement, test, and deploy MCP servers in production.

What Makes MCP Different

Traditional API integrations require AI providers to build custom connectors for every service. This creates an N×M problem: N AI providers times M services equals exponential complexity. MCP solves this with a universal protocol that any AI can speak and any service can implement.

The Three Pillars of MCP

MCP servers expose three types of capabilities to AI agents:

🔧 Tools - Functions that agents can execute

Perform calculations
Query databases
Call external APIs
Manipulate files

📚 Resources - Data sources agents can access

Files and documents
Database contents
API responses
Configuration data

💬 Prompts - Pre-defined interaction templates

Structured queries
Response formats
Workflow templates
Best practices

This trinity of capabilities transforms any application into an AI-ready tool server that any MCP-compatible agent can utilize.

The Architecture: Simple Yet Powerful

At its core, MCP follows a client-server architecture built on JSON-RPC 2.0:

MCP Architecture Diagram

Why JSON-RPC 2.0?

Lightweight and human-readable
Built-in request/response correlation
Standardized error handling
Transport-agnostic design

Every MCP message follows this structure:

{
  "jsonrpc": "2.0",
  "id": "unique-request-id",
  "method": "method-name",
  "params": {
    // method-specific parameters
  }
}

Simple, consistent, predictable—exactly what you want in a protocol.

Transport Protocols: Choose Your Fighter

MCP supports multiple transport protocols, each optimized for different use cases:

🖥️ Standard Input/Output (stdio)

The simplest and most secure transport. The client launches the server as a subprocess and communicates via stdin/stdout.

Perfect for:

Local tools and scripts
Maximum security (no network exposure)
Development and testing
CLI tool integration

Configuration:

{
  "command": "node",
  "args": ["my-mcp-server.js"],
  "env": {
    "API_KEY": "your-key"
  }
}

🌐 Streamable HTTP

The go-to transport for network-accessible MCP servers. A single /mcp endpoint handles all operations, supporting both request-response and streaming.

Perfect for:

Cloud-deployed servers
Multi-tenant services
Production deployments
Cross-network access

Required headers:

POST /mcp HTTP/1.1
Content-Type: application/json
Accept: application/json,text/event-stream
Mcp-Session-Id: session-123

⚠️ Server-Sent Events (Deprecated)

SSE as a standalone transport is deprecated as of November 2024. Use Streamable HTTP instead, which incorporates SSE for streaming responses when needed.

The MCP Handshake: Making First Contact

Every MCP connection begins with a three-step handshake that establishes capabilities and protocol compatibility:

Step 1: Initialize

The client introduces itself and declares its capabilities:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {
    "protocolVersion": "2024-11-05",
    "capabilities": {
      "roots": {
        "listChanged": true
      },
      "sampling": {}
    },
    "clientInfo": {
      "name": "my-ai-agent",
      "version": "1.0.0"
    }
  }
}

Step 2: Server Response

The server responds with its capabilities:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "protocolVersion": "2024-11-05",
    "capabilities": {
      "tools": {
        "listChanged": false
      },
      "resources": {
        "subscribe": true,
        "listChanged": true
      },
      "prompts": {
        "listChanged": false
      }
    },
    "serverInfo": {
      "name": "example-server",
      "version": "1.0.0"
    }
  }
}

Step 3: Initialized Notification

The client confirms it's ready:

{
  "jsonrpc": "2.0",
  "method": "notifications/initialized",
  "params": {}
}

Now the connection is established and both parties know each other's capabilities. Time to get to work!

Core MCP Methods: The Essential Toolkit

Once connected, clients can discover and use server capabilities through these core methods:

Discovering Tools

// Request
{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/list",
  "params": {}
}

// Response
{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "tools": [
      {
        "name": "calculate",
        "description": "Perform mathematical calculations",
        "inputSchema": {
          "type": "object",
          "properties": {
            "expression": {
              "type": "string",
              "description": "Mathematical expression to evaluate"
            }
          },
          "required": ["expression"]
        }
      }
    ]
  }
}

Calling Tools

// Request
{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "tools/call",
  "params": {
    "name": "calculate",
    "arguments": {
      "expression": "2 + 2"
    }
  }
}

// Response
{
  "jsonrpc": "2.0",
  "id": 3,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "4"
      }
    ]
  }
}

Accessing Resources

// List available resources
{
  "jsonrpc": "2.0",
  "id": 4,
  "method": "resources/list",
  "params": {}
}

// Read a specific resource
{
  "jsonrpc": "2.0",
  "id": 5,
  "method": "resources/read",
  "params": {
    "uri": "file:///data/config.json"
  }
}

Building Your First MCP Server

Let's build a simple MCP server that exposes a calculation tool. Here's a complete TypeScript implementation:

import { Server } from '@modelcontextprotocol/sdk/server';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio';

const server = new Server({
  name: 'calculator-server',
  version: '1.0.0',
});

// Define the calculate tool
server.setRequestHandler('tools/list', async () => ({
  tools: [
    {
      name: 'calculate',
      description: 'Evaluate mathematical expressions',
      inputSchema: {
        type: 'object',
        properties: {
          expression: {
            type: 'string',
            description: 'Math expression to evaluate',
          },
        },
        required: ['expression'],
      },
    },
  ],
}));

// Implement the tool
server.setRequestHandler('tools/call', async (request) => {
  if (request.params.name === 'calculate') {
    try {
      // WARNING: eval is dangerous! Use a proper math library in production
      const result = eval(request.params.arguments.expression);
      return {
        content: [
          {
            type: 'text',
            text: result.toString(),
          },
        ],
      };
    } catch (error) {
      return {
        content: [
          {
            type: 'text',
            text: `Error: ${error.message}`,
          },
        ],
        isError: true,
      };
    }
  }

  throw new Error(`Unknown tool: ${request.params.name}`);
});

// Start the server
const transport = new StdioServerTransport();
server.connect(transport);

Save this as calculator-server.ts and run it with:

npx tsx calculator-server.ts

Your MCP server is now ready to accept connections!

Testing MCP Servers with curl

Testing is crucial for MCP development. Here's how to test any MCP server using curl:

Testing HTTP Transport

1. Initialize the connection:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json,text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {
        "name": "curl-test",
        "version": "1.0"
      }
    }
  }'

2. Send initialized notification:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "notifications/initialized",
    "params": {}
  }'

3. List available tools:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/list",
    "params": {}
  }'

4. Call a tool:

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "calculate",
      "arguments": {
        "expression": "42 * 10"
      }
    }
  }'

Testing with Session Management

Some servers require session management. Here's how to handle it:

# Create a session
SESSION_ID="test-session-$(date +%s)"

# Initialize with session
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json,text/event-stream" \
  -H "Mcp-Session-Id: $SESSION_ID" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {
        "name": "curl-test",
        "version": "1.0"
      }
    }
  }'

# Use the same session for subsequent requests
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -H "Mcp-Session-Id: $SESSION_ID" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/list",
    "params": {}
  }'

# Clean up the session when done
curl -X DELETE http://localhost:8000/mcp \
  -H "Mcp-Session-Id: $SESSION_ID"

Testing stdio Transport

For stdio servers, use echo and pipes:

# Simple one-shot test
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' | node calculator-server.js

# Interactive testing with named pipes
mkfifo /tmp/mcp_in /tmp/mcp_out
node calculator-server.js < /tmp/mcp_in > /tmp/mcp_out &

# Send requests
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' > /tmp/mcp_in

# Read responses
cat /tmp/mcp_out

Error Handling: When Things Go Wrong

MCP uses standard JSON-RPC error codes plus custom ones for protocol-specific issues:

Standard Error Codes

Code	Meaning	When to Use
-32700	Parse error	Invalid JSON
-32600	Invalid request	Missing required fields
-32601	Method not found	Unknown method name
-32602	Invalid params	Wrong parameter types/values
-32603	Internal error	Server-side failure

Error Response Format

{
  "jsonrpc": "2.0",
  "id": "request-id",
  "error": {
    "code": -32601,
    "message": "Method not found",
    "data": {
      "method": "unknown/method",
      "availableMethods": ["tools/list", "tools/call"]
    }
  }
}

Real-World MCP Servers

The MCP ecosystem is growing rapidly with servers for popular services:

Official MCP Servers

GitHub MCP Server

Repository operations (create, clone, push)
Issue and PR management
Code search and navigation

Slack MCP Server

Send messages
Read channels
User management

Postgres MCP Server

Execute queries
Schema exploration
Transaction management

Google Drive MCP Server

File operations
Search documents
Share management

Community Servers

The community has built MCP servers for everything from weather APIs to cryptocurrency exchanges. Check out the comprehensive list at: https://github.com/cyanheads/model-context-protocol-resources

Security Best Practices

MCP servers handle sensitive operations. Follow these security guidelines:

1. Authentication Always

Never expose an MCP server without authentication:

server.setRequestHandler('initialize', async (request, context) => {
  // Validate auth token
  const token = context.headers['authorization'];
  if (!validateToken(token)) {
    throw new Error('Unauthorized');
  }

  // Continue with initialization
  return { /* ... */ };
});

2. Input Validation

Always validate and sanitize inputs:

server.setRequestHandler('tools/call', async (request) => {
  const { name, arguments: args } = request.params;

  // Validate tool name
  if (!ALLOWED_TOOLS.includes(name)) {
    throw new Error(`Unknown tool: ${name}`);
  }

  // Validate arguments
  const schema = TOOL_SCHEMAS[name];
  if (!validateSchema(args, schema)) {
    throw new Error('Invalid arguments');
  }

  // Safe to proceed
  return executeTool(name, args);
});

3. Rate Limiting

Prevent abuse with rate limiting:

const rateLimiter = new Map();

server.setRequestHandler('tools/call', async (request, context) => {
  const clientId = context.sessionId || context.ip;
  const limit = rateLimiter.get(clientId) || 0;

  if (limit > MAX_REQUESTS_PER_MINUTE) {
    throw new Error('Rate limit exceeded');
  }

  rateLimiter.set(clientId, limit + 1);

  // Process request
});

4. Secure Transport

For production, always use:

HTTPS for HTTP transport
Environment variables for secrets
Proper certificate validation
Network isolation where possible

Troubleshooting Common Issues

"Method not found" Error

Verify the method name is correct
Check protocol version compatibility
Ensure server implements the capability

Connection Refused

Verify server is running
Check correct port and endpoint
Ensure firewall allows connection

Invalid JSON-RPC Format

Always include "jsonrpc": "2.0"
Include id for requests (not notifications)
Use correct parameter structure

Session Errors

Some servers require session management
Include Mcp-Session-Id header consistently
Clean up sessions with DELETE request

The MCP Ecosystem in 2025

MCP adoption has exploded in 2025 with major players embracing the standard:

OpenAI - Native MCP support in GPT models Google - Gemini models speak MCP natively Microsoft - Azure AI services include MCP connectors Amazon - Bedrock supports MCP servers

This industry-wide adoption means:

Write once, use everywhere
Reduced integration complexity
Faster time to market
Lower maintenance costs

Monitoring MCP Servers with AgentFlare

While MCP standardizes connections, you still need visibility into what's happening. AgentFlare provides comprehensive observability for MCP servers:

Complete Request Tracking

Every tool call logged and analyzed
Request/response payloads captured
Latency and error tracking
Cost attribution per operation

Real-Time Monitoring

Live dashboard showing MCP traffic
Alert on errors or performance issues
Track usage patterns
Identify optimization opportunities

Simple Integration

Just point your MCP client through AgentFlare's proxy:

const client = new Client({
  transport: new HTTPTransport({
    baseUrl: "https://your-workspace.agentflare.com/proxy",
    headers: {
      "Authorization": `Bearer ${API_KEY}`,
      "X-Target-Server": "your-mcp-server"
    }
  })
});

Now every MCP interaction is automatically monitored, giving you the insights needed to optimize performance, control costs, and debug issues.

Getting Started with MCP

Ready to implement MCP? Here's your action plan:

1. Choose Your Use Case

Start with a simple, well-defined tool that would benefit from AI integration.

2. Select a Transport

stdio for local development
HTTP for production deployment

3. Implement Core Methods

Start with initialize and tools/list
Add tools/call for your specific tools
Expand to resources and prompts as needed

4. Test Thoroughly

Use curl for manual testing
Implement automated tests
Try the MCP Inspector tool

5. Deploy with Monitoring

Use AgentFlare for observability
Set up alerts for errors
Track usage and costs

The Future of MCP

MCP is more than a protocol—it's becoming the foundation for AI agent infrastructure. As the ecosystem matures, we're seeing:

Richer capabilities: Video, audio, and real-time streaming
Better tooling: IDEs with MCP support, automated testing frameworks
Marketplace emergence: Discover and share MCP servers
Enterprise features: Advanced auth, compliance, audit trails

The companies that master MCP today will have a significant advantage as AI agents become central to business operations.

Conclusion: MCP Changes Everything

The Model Context Protocol represents a fundamental shift in how we think about AI integration. Instead of building custom connectors for every AI provider, we can now build once and support them all. Instead of proprietary protocols, we have an open standard. Instead of fragmentation, we have unity.

For developers, MCP means less time on integration and more time on innovation. For businesses, it means faster deployment and lower costs. For the industry, it means a more robust and interoperable AI ecosystem.

The USB-C moment for AI is here. The question isn't whether to adopt MCP—it's how quickly you can get started.

Resources and Next Steps

Official Resources:

MCP Specification - Complete protocol documentation
MCP Inspector - Debugging tool
Example Servers - Reference implementations

AgentFlare Resources:

MCP Proxy Setup Guide - Configure AgentFlare for MCP monitoring
Getting Started - Begin monitoring your MCP servers
Tool Server Observability - Understanding the bigger picture

Community:

MCP Resources List - Community servers and tools
MCP Discord - Join the discussion

Ready to monitor your MCP servers? Start with AgentFlare for complete observability of your AI tool infrastructure.