MCP Server - Model Context Protocol Server

The BoxLang AI Module provides a complete MCP (Model Context Protocol) server implementation that allows you to expose tools, resources, and prompts to AI clients.

📋 Table of Contents

🎯 What is an MCP Server?

An MCP Server is a service that exposes capabilities to AI clients using the standardized Model Context Protocol. It enables:

🔧 Expose Tools: Register functions that AI clients can invoke
📚 Serve Resources: Provide documents and data to AI clients
💬 Offer Prompts: Define reusable prompt templates
🌐 HTTP & STDIO Transports: Expose your MCP server via web or command-line

🏗️ MCP Architecture

🚀 Transport Types

BoxLang AI provides two transport mechanisms for MCP servers:

🔄 Transport Flow Comparison

🌐 HTTP Transport (Web)

Best for: Web applications, REST APIs, browser-based clients

Entry Point: Automatic via convention or manual endpoint creation

Automatic (Convention):

// Access the built-in endpoint
POST http://localhost/~bxai/mcp.bxm
POST http://localhost/~bxai/mcp.bxm?server=myApp
POST http://localhost/~bxai/mcp.bxm/myApp

Manual (Custom Endpoint):

// api/my-mcp.bxm
<bx:script>
import bxModules.bxai.models.mcp.MCPRequestProcessor
MCPRequestProcessor::startHttp()
</bx:script>

Features:

✅ CORS support with wildcard patterns
✅ Body size limits
✅ API key authentication
✅ HTTP Basic Auth
✅ Security headers
✅ Server-Sent Events (SSE) streaming
✅ Discovery endpoint (GET)

🖥️ STDIO Transport (Command-Line)

Best for: Desktop applications, CLI tools, IDE integrations, local AI assistants

Entry Point: Create a script entry point

Example Script (mcp-stdio.bxs):

import bxModules.bxai.models.mcp.MCPRequestProcessor

// Register your server and tools
MCPServer( "myApp" )
    .registerTool( aiTool( "echo", "Echo tool", ( msg ) => msg ) )

// Get server name from CLI args
serverName = getSystemSetting( "MCP_SERVER_NAME", "default" )
parsedArgs = cliGetArgs()

if ( structKeyExists( parsedArgs.options, "server" ) ) {
    serverName = parsedArgs.options.server
}

// Start STDIO transport (blocks until shutdown)
MCPRequestProcessor::startStdio( serverName )

Run:

# Start with default server
boxlang mcp-stdio.bxs

# Start with specific server
boxlang mcp-stdio.bxs --server myApp

# Or via environment variable
MCP_SERVER_NAME=myApp boxlang mcp-stdio.bxs

Features:

✅ JSON-RPC over STDIN/STDOUT
✅ Line-based protocol
✅ Graceful shutdown signal
✅ Process lifecycle management
❌ No HTTP headers/CORS (not needed)
❌ No status codes (JSON-RPC error codes used)

Communication:

# Input (STDIN)
{"jsonrpc":"2.0","method":"tools/list","params":{},"id":1}

# Output (STDOUT)
{"jsonrpc":"2.0","result":{"tools":[...]},"id":1}

# Shutdown
{"jsonrpc":"2.0","method":"shutdown","params":{},"id":99}

📖 Complete Example: See examples/mcp/09-mcp-stdio.bxs for a full working STDIO entry point with argument parsing.

🎯 Choosing a Transport

Scenario

Recommended Transport

Web application with browser clients

HTTP

REST API for external services

HTTP

Desktop AI assistant (Claude Desktop, etc.)

STDIO

VS Code extension / IDE integration

STDIO

Command-line tool

STDIO

Docker container as MCP service

Both (HTTP for API, STDIO for container tools)

Testing/Development

HTTP (easier to test with curl)

Quick Start

1. Register Tools

// Application.bx
class {

    function onApplicationStart() {
        // Get or create an MCP server instance
        MCPServer( "myApp" )
            .setDescription( "My Application MCP Server" )
            .setVersion( "1.0.0" )
            .registerTool(
                aiTool( "search", "Search for documents", ( query ) => {
                    return searchService.search( query )
                } )
            )
            .registerTool(
                aiTool( "calculate", "Perform calculations", ( expression ) => {
                    return evaluate( expression )
                } )
            )
    }

}

2. Access the MCP Endpoint

The module provides a built-in endpoint at public/mcp.bxm:

POST http://localhost/~bxai/mcp.bxm

Specify a server using either a query parameter or URL segment:

# Using query parameter
POST http://localhost/~bxai/mcp.bxm?server=myApp

# Using URL segment
POST http://localhost/~bxai/mcp.bxm/myApp

Custom URL Entry Points

You can create your own MCP server endpoints at any URL by importing and rendering the MCPRequestProcessor:

// api/mcp-endpoint.bxm
<bx:script>
import bxModules.bxai.models.mcp.MCPRequestProcessor
MCPRequestProcessor::startHttp()
</bx:script>

This creates a fully functional MCP endpoint at your custom location:

POST http://localhost/api/mcp-endpoint.bxm?server=myApp

Use Cases:

Custom routing: Place MCP endpoints under your API structure (/api/v1/mcp, /admin/mcp, etc.)
Security: Put endpoints behind authentication middleware or custom security rules
Multiple entry points: Create different endpoints for different purposes (public API, admin tools, etc.)
Framework integration: Integrate MCP servers into existing URL routing schemes

Example - Secured Admin Endpoint:

// admin/mcp-admin.bxm
<bx:script>
// Check admin authentication
if ( !session.isAdmin ) {
    writeOutput( serializeJSON({
        jsonrpc: "2.0",
        error: {
            code: -32000,
            message: "Unauthorized"
        },
        id: null
    }) )
    abort;
}

// Render MCP processor
import bxModules.bxai.models.mcp.MCPRequestProcessor
MCPRequestProcessor::startHttp()
</bx:script>

Example - API Versioned Endpoint:

// api/v2/ai/mcp.bxm
<bx:script>
// Set API version header
request.setHeader( "X-API-Version", "2.0" )

// Render MCP processor
import bxModules.bxai.models.mcp.MCPRequestProcessor
MCPRequestProcessor::startHttp()
</bx:script>

The processor automatically:

✅ Extracts server name from query parameter (?server=name) or URL segment
✅ Handles JSON-RPC 2.0 request parsing
✅ Routes to the correct MCP server instance
✅ Returns properly formatted JSON-RPC responses
✅ Fires all MCP events (onMCPRequest, onMCPResponse, onMCPError)

Server Configuration

Creating a Server

// Get or create a server instance (singleton by name)
server = MCPServer( "myApp" )

// Multiple servers for different purposes
apiServer = MCPServer( "api" )
adminServer = MCPServer( "admin" )

Configure Description and Version

server = MCPServer( "myApp" )
    .setDescription( "My Application MCP Server" )
    .setVersion( "2.0.0" )

Get Server Info

info = server.getServerInfo()
// { name: "myApp", version: "2.0.0" }

Basic Authentication 🔒

Protect your MCP server with HTTP Basic Authentication:

// Configure basic auth credentials
server = MCPServer( "myApp" )
    .withBasicAuth( "admin", "secretPassword123" )
    .registerTool( myTool )

How it works:

Credentials are verified before any request processing
Returns 401 Unauthorized with WWW-Authenticate header if credentials are invalid
Uses standard HTTP Basic Authentication (base64-encoded username:password)
Zero performance overhead when not configured
Fluent API for easy configuration

Example - Secured Admin Server:

// Application.bx
function onApplicationStart() {
    MCPServer( "admin" )
        .withBasicAuth( "admin", application.adminPassword )
        .setDescription( "Admin MCP Server - Requires Authentication" )
        .registerTool( adminTool )
        .registerTool( systemTool )
}

Making authenticated requests:

# Using curl with basic auth
curl -X POST http://localhost/~bxai/mcp.bxm?server=admin \
  -u admin:secretPassword123 \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":"1"}'

# Or with Authorization header
curl -X POST http://localhost/~bxai/mcp.bxm?server=admin \
  -H "Authorization: Basic YWRtaW46c2VjcmV0UGFzc3dvcmQxMjM=" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":"1"}'

Check if authentication is enabled:

if ( server.hasBasicAuth() ) {
    writeOutput( "This server requires authentication" )
}

Security Best Practices:

✅ Always use HTTPS in production to prevent credential interception
✅ Store passwords in environment variables or encrypted configuration
✅ Use strong, unique passwords for each server
✅ Combine with CORS settings for additional security
✅ Consider rotating credentials periodically
✅ Log authentication failures for security monitoring

Example with environment variables:

// Application.bx
function onApplicationStart() {
    var adminUser = getEnv( "MCP_ADMIN_USER" )
    var adminPass = getEnv( "MCP_ADMIN_PASS" )

    if ( len( adminUser ) && len( adminPass ) ) {
        MCPServer( "admin" )
            .withBasicAuth( adminUser, adminPass )
            .registerTool( adminTool )
    }
}

CORS Configuration 🌐

Configure Cross-Origin Resource Sharing (CORS) to control which origins can access your MCP server:

// Allow a single origin
server = MCPServer( "myApp" )
    .withCors( "https://example.com" )

// Allow multiple origins
server = MCPServer( "myApp" )
    .withCors( [ "https://example.com", "https://app.example.com" ] )

// Allow all origins (use with caution!)
server = MCPServer( "myApp" )
    .withCors( "*" )

// Wildcard subdomain matching
server = MCPServer( "myApp" )
    .withCors( [ "*.example.com", "https://trusted.org" ] )

Wildcard Pattern Matching:

The CORS implementation supports wildcard patterns for flexible domain matching:

*.example.com - Matches any subdomain: app.example.com, api.example.com, admin.example.com
* - Matches all origins (returns the requesting origin in Access-Control-Allow-Origin)
Exact matches: https://example.com - Only matches exactly this origin

Dynamic CORS Management:

// Add origins incrementally
server = MCPServer( "myApp" )
    .addCorsOrigin( "https://example.com" )
    .addCorsOrigin( "https://app.example.com" )

// Get configured origins
origins = server.getCorsAllowedOrigins()
// Returns: [ "https://example.com", "https://app.example.com" ]

// Check if an origin is allowed
if ( server.isCorsAllowed( "https://example.com" ) ) {
    writeOutput( "Origin is allowed" )
}

How CORS Works:

When a browser makes a cross-origin request:

Browser sends Origin header with the request
Server checks if origin matches allowed patterns
If allowed, server responds with Access-Control-Allow-Origin header
Server also includes CORS headers in OPTIONS preflight responses

CORS Headers Set by Server:

Access-Control-Allow-Origin - Allowed origin (dynamic or *)
Access-Control-Allow-Methods - GET, POST, OPTIONS
Access-Control-Allow-Headers - Content-Type, Authorization, X-API-Key
Access-Control-Max-Age - 86400 (24 hours)

Security Best Practices:

✅ Avoid * in production - Specify exact origins or wildcard patterns
✅ Use HTTPS origins - Always prefer secure origins
✅ Combine with authentication - CORS doesn't replace authentication
✅ Review periodically - Remove unused origins
✅ Test preflight requests - Verify OPTIONS requests work correctly

Example - Multi-Environment Setup:

// Application.bx
function onApplicationStart() {
    var allowedOrigins = []

    // Add origins based on environment
    if ( getEnv( "ENVIRONMENT" ) == "production" ) {
        allowedOrigins = [
            "https://app.example.com",
            "https://admin.example.com"
        ]
    } else {
        // Development - allow local testing
        allowedOrigins = [
            "http://localhost:3000",
            "http://localhost:8080",
            "*.ngrok.io"  // For tunneling/testing
        ]
    }

    MCPServer( "api" )
        .withCors( allowedOrigins )
        .registerTool( myTool )
}

Request Body Size Limits 📏

Protect your server from large payloads by setting request body size limits:

// Limit to 1MB (in bytes)
server = MCPServer( "myApp" )
    .withBodyLimit( 1048576 )

// Limit to 500KB
server = MCPServer( "myApp" )
    .withBodyLimit( 500 * 1024 )

// Unlimited (default: 0)
server = MCPServer( "myApp" )
    .withBodyLimit( 0 )

How it works:

Server checks len(requestBody) before processing
If body exceeds limit, returns 413 Payload Too Large error
Default is 0 (unlimited)
Limit applies to entire JSON-RPC request body

Error Response (413):

{
  "jsonrpc": "2.0",
  "error": {
    "code": -32000,
    "message": "Request body too large (max: 1048576 bytes)"
  },
  "id": null
}

Get current limit:

maxSize = server.getMaxRequestBodySize()
// Returns: 1048576 (or 0 for unlimited)

Use Cases:

Public APIs - Prevent abuse from extremely large payloads
Resource constraints - Match server memory/processing limits
Tool-specific limits - Different servers can have different limits
DoS prevention - Basic protection against payload attacks

Example - Tiered Limits:

// Public API - strict limits
MCPServer( "public" )
    .withBodyLimit( 100 * 1024 )  // 100KB
    .registerTool( publicTool )

// Internal API - generous limits
MCPServer( "internal" )
    .withBodyLimit( 10 * 1024 * 1024 )  // 10MB
    .withBasicAuth( "admin", "secret" )
    .registerTool( adminTool )

// Data import - unlimited
MCPServer( "import" )
    .withBodyLimit( 0 )  // No limit
    .withBasicAuth( "importer", "secret" )
    .registerTool( importTool )

Custom API Key Validation 🔑

Implement custom API key authentication logic with provider callbacks:

// Simple API key validation
server = MCPServer( "myApp" )
    .withApiKeyProvider( ( apiKey, requestData ) => {
        return apiKey == "my-secret-key-12345"
    } )

// Database lookup
server = MCPServer( "myApp" )
    .withApiKeyProvider( ( apiKey, requestData ) => {
        var user = userService.findByApiKey( apiKey )
        return !isNull( user ) && user.isActive
    } )

// Complex validation with rate limiting
server = MCPServer( "myApp" )
    .withApiKeyProvider( ( apiKey, requestData ) => {
        var key = apiKeyService.validate( apiKey )
        if ( isNull( key ) ) return false

        // Check rate limits
        if ( rateLimiter.isExceeded( key.userId ) ) {
            throw( "Rate limit exceeded", "RateLimitError" )
        }

        // Log usage
        auditLog.record( key.userId, requestData.method )

        return true
    } )

Provider Function Signature:

function apiKeyProvider(
    required string apiKey,      // The API key from request
    required struct requestData  // Request context
) {
    // Return true to allow, false to deny
    // Throw exception for custom error messages
    return true
}

Request Data Struct:

The requestData argument contains:

method - MCP method being called (e.g., "tools/list")
serverName - Name of the MCP server
body - Full request body as string
Any other request metadata

API Key Extraction:

The server automatically extracts API keys from:

X-API-Key header
Authorization: Bearer <token> header

How it works:

Server checks for API key in headers
If found, calls your provider function with key and request data
If provider returns false or throws, returns 401 Unauthorized
If provider returns true, request proceeds normally

Error Response (401):

{
  "jsonrpc": "2.0",
  "error": {
    "code": -32000,
    "message": "Invalid API key"
  },
  "id": null
}

Check if provider is configured:

if ( server.hasApiKeyProvider() ) {
    writeOutput( "API key validation is enabled" )
}

Making authenticated requests:

# Using X-API-Key header
curl -X POST http://localhost/~bxai/mcp.bxm?server=myApp \
  -H "X-API-Key: my-secret-key-12345" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":"1"}'

# Using Bearer token
curl -X POST http://localhost/~bxai/mcp.bxm?server=myApp \
  -H "Authorization: Bearer my-secret-key-12345" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":"1"}'

Advanced Example - Multi-Tenant with Usage Tracking:

// Application.bx
function onApplicationStart() {
    MCPServer( "api" )
        .withApiKeyProvider( ( apiKey, requestData ) => {
            // Validate and get tenant
            var tenant = tenantService.validateApiKey( apiKey )
            if ( isNull( tenant ) ) return false

            // Check feature access
            if ( !tenant.hasFeature( requestData.method ) ) {
                throw( "Feature not available in your plan", "AccessDenied" )
            }

            // Store tenant context for later use
            request.tenantId = tenant.id
            request.tenantName = tenant.name

            // Track usage
            usageTracker.record(
                tenantId: tenant.id,
                method: requestData.method,
                timestamp: now()
            )

            return true
        } )
        .registerTool( myTool )
}

Combining Security Features:

You can use multiple security features together:

server = MCPServer( "enterprise" )
    // CORS - restrict origins
    .withCors( [ "https://app.example.com", "*.example.com" ] )
    // Body limits - prevent abuse
    .withBodyLimit( 1048576 )  // 1MB
    // API keys - identify clients
    .withApiKeyProvider( validateApiKey )
    // Basic auth - admin access
    .withBasicAuth( "admin", "secret" )
    // Tools
    .registerTool( toolOne )
    .registerTool( toolTwo )

Security Priority:

Body size check - Happens first (reject oversized payloads immediately)
CORS validation - Checks Origin header against allowed patterns
Basic authentication - HTTP Basic Auth credentials check
API key validation - Custom provider callback execution
Request processing - Only if all checks pass

Security Headers 🛡️

The MCP server automatically includes industry-standard security headers in all responses:

Headers Included:

X-Content-Type-Options: nosniff - Prevents MIME type sniffing
X-Frame-Options: DENY - Prevents clickjacking attacks
X-XSS-Protection: 1; mode=block - Enables XSS filtering
Referrer-Policy: strict-origin-when-cross-origin - Controls referrer information
Content-Security-Policy: default-src 'none'; frame-ancestors 'none' - Restricts resource loading
Strict-Transport-Security: max-age=31536000; includeSubDomains - Forces HTTPS (when applicable)
Permissions-Policy: geolocation=(), microphone=(), camera=() - Disables sensitive browser features

Automatic Application:

Security headers are added to:

✅ Successful responses (200 OK)
✅ Error responses (400, 401, 404, 413, 500)
✅ CORS preflight responses (OPTIONS)

No configuration needed - these headers are applied automatically to enhance security posture.

Tool Registration

Register a Single Tool

server = MCPServer( "myApp" )
    .registerTool(
        aiTool( "getWeather", "Get current weather for a location", ( location ) => {
            return weatherService.getCurrent( location )
        } )
        .describeArg( "location", "City name or coordinates" )
    )

Register Multiple Tools

tools = [
    aiTool( "search", "Search documents", searchHandler ),
    aiTool( "translate", "Translate text", translateHandler ),
    aiTool( "summarize", "Summarize text", summarizeHandler )
]

server = MCPServer( "myApp" )
    .registerTools( tools )

Check and Retrieve Tools

// Check if a tool exists
exists = server.hasTool( "search" )

// Get a specific tool
tool = server.getTool( "search" )

// Get tool count
count = server.getToolCount()

// Get all tools
tools = server.getTools()

List Tools (MCP Format)

// Returns array formatted for MCP protocol
toolsList = server.listTools()
// [
//     {
//         name: "search",
//         description: "Search documents",
//         inputSchema: {
//             type: "object",
//             properties: { ... },
//             required: [ ... ]
//         }
//     }
// ]

Unregister Tools

// Remove a specific tool
server.unregisterTool( "oldTool" )

// Remove all tools
server.clearTools()

Annotation-Based Discovery

The MCP server can automatically discover and register tools, resources, and prompts from annotated methods using the scan() method.

Scan for Annotations

// Scan a class file
MCPServer( "myApp" ).scan( "/path/to/MyTools.bx" )

// Scan a directory (recursively scans all .bx files)
MCPServer( "myApp" ).scan( "/path/to/tools/" )

@mcpTool Annotation

class {

    /**
     * Search for documents
     * @query The search query
     */
    @mcpTool
    function search( required string query ) {
        return searchService.find( query )
    }

    /**
     * @query The query to search
     */
    @mcpTool( "Search for documents in the knowledge base" )
    function searchDocs( required string query ) {
        return docService.search( query )
    }

    @mcpTool( { name: "calculator", description: "Perform calculations", version: "2.0.0" } )
    function calculate( required string expression ) {
        return evaluate( expression )
    }

}

Annotation formats:

@mcpTool - Name from method name, description from function hint, version defaults to 1.0.0
@mcpTool( "Description" ) - Name from method name, custom description
@mcpTool( { name: "...", description: "...", version: "..." } ) - All custom values

@mcpResource Annotation

class {

    /**
     * Returns the README file
     */
    @mcpResource
    function readme() {
        return fileRead( "/readme.md" )
    }

    @mcpResource( "API documentation for developers" )
    function apiDocs() {
        return generateApiDocs()
    }

    @mcpResource( { uri: "config://app", name: "App Config", description: "Application settings", mimeType: "application/json" } )
    function getConfig() {
        return application.settings
    }

}

Annotation formats:

@mcpResource - URI and name from method name, description from function hint
@mcpResource( "Description" ) - URI and name from method name, custom description
@mcpResource( { uri: "...", name: "...", description: "...", mimeType: "..." } ) - All custom values

@mcpPrompt Annotation

class {

    /**
     * Generate a greeting message
     * @name The person's name
     */
    @mcpPrompt
    function greeting( required string name ) {
        return [
            { role: "system", content: "You are a friendly assistant." },
            { role: "user", content: "Say hello to #name#" }
        ]
    }

    @mcpPrompt( "Generate code based on a description" )
    function codeGen( required string description, string language = "java" ) {
        return [
            { role: "system", content: "You are a code generator for #language#." },
            { role: "user", content: description }
        ]
    }

    @mcpPrompt( { name: "reviewer", description: "Code review prompt", arguments: [ { name: "code", required: true } ] } )
    function reviewCode( required string code ) {
        return [
            { role: "system", content: "Review this code for issues." },
            { role: "user", content: code }
        ]
    }

}

Annotation formats:

@mcpPrompt - Name from method name, description from function hint
@mcpPrompt( "Description" ) - Name from method name, custom description
@mcpPrompt( { name: "...", description: "...", arguments: [...] } ) - All custom values

Resource Registration

Resources provide access to documents and data:

Register a Resource

server = MCPServer( "myApp" )
    .registerResource(
        uri: "docs://readme",
        name: "README",
        description: "Project documentation",
        mimeType: "text/markdown",
        handler: () => {
            return fileRead( expandPath( "/readme.md" ) )
        }
    )

Register Dynamic Resources

// Database content
server.registerResource(
    uri: "db://users",
    name: "User List",
    description: "Current user data",
    mimeType: "application/json",
    handler: () => {
        return userService.getAllUsers()
    }
)

// Configuration
server.registerResource(
    uri: "config://app",
    name: "App Configuration",
    description: "Application settings",
    mimeType: "application/json",
    handler: () => {
        return application.settings
    }
)

List and Read Resources

// List available resources
resources = server.listResources()

// Check if resource exists
exists = server.hasResource( "docs://readme" )

// Read a resource
content = server.readResource( "docs://readme" )
// {
//     contents: [
//         {
//             uri: "docs://readme",
//             mimeType: "text/markdown",
//             text: "# Readme content..."
//         }
//     ]
// }

Manage Resources

// Remove a resource
server.unregisterResource( "docs://readme" )

// Clear all resources
server.clearResources()

Prompt Registration

Prompts provide reusable prompt templates:

Register a Prompt

server = MCPServer( "myApp" )
    .registerPrompt(
        name: "codeReview",
        description: "Code review prompt template",
        args: [
            { name: "language", description: "Programming language", required: true },
            { name: "code", description: "Code to review", required: true }
        ],
        handler: ( args ) => {
            return [
                {
                    role: "system",
                    content: "You are a code reviewer specializing in #args.language#."
                },
                {
                    role: "user",
                    content: "Please review this code:\n\n```#args.language#\n#args.code#\n```"
                }
            ]
        }
    )

List and Get Prompts

// List available prompts
prompts = server.listPrompts()

// Check if prompt exists
exists = server.hasPrompt( "codeReview" )

// Get a prompt with arguments
result = server.getPrompt( "codeReview", {
    language: "java",
    code: "public void test() {}"
} )
// {
//     description: "Code review prompt template",
//     messages: [
//         { role: "system", content: { type: "text", text: "..." } },
//         { role: "user", content: { type: "text", text: "..." } }
//     ]
// }

Handling MCP Requests

Direct Request Handling

// Handle a JSON-RPC request
request = {
    "jsonrpc": "2.0",
    "method": "tools/list",
    "id": "1"
}

response = server.handleRequest( request )
// {
//     jsonrpc: "2.0",
//     result: { tools: [ ... ] },
//     id: "1"
// }

Handle JSON String

jsonRequest = '{"jsonrpc":"2.0","method":"tools/call","id":"1","params":{"name":"search","arguments":{"query":"test"}}}'

response = server.handleRequest( jsonRequest )

Supported Methods

Method

Description

initialize

Get server capabilities and info

tools/list

List available tools

tools/call

Invoke a tool

resources/list

List available resources

resources/read

Read a resource

prompts/list

List available prompts

prompts/get

Get a prompt with arguments

ping

Health check

HTTP Endpoint (mcp.bxm)

The module includes a pre-built HTTP endpoint at public/mcp.bxm:

Discovery (GET)

curl http://localhost/~bxai/mcp.bxm

Returns server capabilities:

{
    "jsonrpc": "2.0",
    "result": {
        "protocolVersion": "2024-11-05",
        "capabilities": {
            "tools": {},
            "resources": {},
            "prompts": {}
        },
        "serverInfo": {
            "name": "default",
            "version": "1.0.0"
        }
    }
}

JSON-RPC Requests (POST)

curl -X POST http://localhost/~bxai/mcp.bxm \
    -H "Content-Type: application/json" \
    -d '{"jsonrpc":"2.0","method":"tools/list","id":"1"}'

Multiple Servers

Specify the server using either a query parameter or URL segment:

# Using query parameter
curl http://localhost/~bxai/mcp.bxm?server=api
curl http://localhost/~bxai/mcp.bxm?server=admin

# Using URL segment
curl http://localhost/~bxai/mcp.bxm/api
curl http://localhost/~bxai/mcp.bxm/admin

CORS Support

The endpoint includes CORS headers for cross-origin requests.

Static Server Management

Check Server Existence

// Check if a server exists
exists = bxModules.bxai.models.mcp.MCPServer::hasInstance( "myApp" )

Get All Server Names

names = bxModules.bxai.models.mcp.MCPServer::getInstanceNames()
// [ "default", "api", "admin" ]

Remove a Server

wasRemoved = bxModules.bxai.models.mcp.MCPServer::removeInstance( "oldApp" )

Clear All Servers

bxModules.bxai.models.mcp.MCPServer::clearAllInstances()

Complete Example

Application Setup

// Application.bx
class {

    function onApplicationStart() {
        // Create the MCP server
        var server = MCPServer( "myApp" )
            .setDescription( "My Application API" )
            .setVersion( "1.0.0" )

        // Register tools
        server.registerTool(
            aiTool( "searchProducts", "Search product catalog", ( query, category ) => {
                return productService.search(
                    query: arguments.query,
                    category: arguments.category ?: ""
                )
            } )
            .describeArg( "query", "Search keywords" )
            .describeArg( "category", "Optional category filter" )
        )

        server.registerTool(
            aiTool( "getOrderStatus", "Get order status", ( orderId ) => {
                return orderService.getStatus( arguments.orderId )
            } )
            .describeArg( "orderId", "The order ID to check" )
        )

        // Register resources
        server.registerResource(
            uri: "catalog://products",
            name: "Product Catalog",
            description: "Full product listing",
            mimeType: "application/json",
            handler: () => productService.getAll()
        )

        server.registerResource(
            uri: "docs://api",
            name: "API Documentation",
            description: "API reference documentation",
            mimeType: "text/markdown",
            handler: () => fileRead( expandPath( "/docs/api.md" ) )
        )

        // Register prompts
        server.registerPrompt(
            name: "productRecommendation",
            description: "Get product recommendations",
            args: [
                { name: "preferences", description: "User preferences", required: true }
            ],
            handler: ( args ) => [
                {
                    role: "system",
                    content: "You are a product recommendation assistant."
                },
                {
                    role: "user",
                    content: "Based on these preferences, suggest products: #args.preferences#"
                }
            ]
        )

        return true
    }

    function onApplicationEnd() {
        // Clean up
        bxModules.bxai.models.mcp.MCPServer::removeInstance( "myApp" )
    }

}

Connecting from AI Client

// Use the MCP client to connect to the server
client = MCP( "http://localhost/~bxai/mcp.bxm?server=myApp" )

// List available tools
tools = client.listTools()

// Invoke a tool
result = client.send( "searchProducts", {
    query: "laptop",
    category: "electronics"
} )

if ( result.getSuccess() ) {
    products = result.getData()
    writeOutput( "Found #arrayLen( products )# products" )
}

Events & Interception 🎯

The MCP Server fires custom events during its lifecycle, allowing you to add custom logging, monitoring, alerting, and integration with other systems.

📡 Transport Agnostic: All MCP events work identically for both HTTP and STDIO transports. The same event handlers will be called regardless of how clients connect to your MCP server.

Available Events

`onMCPServerCreate`

Fired when a new MCP server instance is created.

function onMCPServerCreate( event, interceptData ) {
    // interceptData contains:
    // - server: The newly created MCPServer instance
    // - name: The server name
    writeLog( "MCP server created: #interceptData.name#" )
}

`onMCPServerRemove`

Fired when an MCP server instance is removed from the registry.

function onMCPServerRemove( event, interceptData ) {
    // interceptData contains:
    // - name: The server name that was removed
    writeLog( "MCP server removed: #interceptData.name#" )
}

`onMCPRequest`

Fired before processing an incoming MCP request.

function onMCPRequest( event, interceptData ) {
    // interceptData contains:
    // - server: The MCPServer instance
    // - requestData: The parsed request { id, method, params }

    writeLog(
        type: "information",
        file: "mcp-requests",
        text: "Request: #interceptData.requestData.method# (ID: #interceptData.requestData.id#)"
    )
}

`onMCPResponse`

Fired after processing an MCP request and before returning the response.

function onMCPResponse( event, interceptData ) {
    // interceptData contains:
    // - server: The MCPServer instance
    // - requestData: The original request
    // - responseData: The response being returned
    // - success: Whether the request was successful
    // - responseTime: Time taken in milliseconds

    if ( !interceptData.success ) {
        writeLog(
            type: "warning",
            file: "mcp-failures",
            text: "Failed request: #interceptData.requestData.method#"
        )
    }
}

`onMCPError` ⚠️

Fired when an exception occurs during MCP server operations.

function onMCPError( event, interceptData ) {
    // interceptData contains:
    // - server: The MCPServer instance
    // - context: Where the error occurred ("handleRequest", "scanClass", etc.)
    // - exception: The exception object with message, detail, stacktrace
    // - Additional context-specific fields

    var exception = interceptData.exception
    var context = interceptData.context

    // Log detailed error
    writeLog(
        type: "error",
        file: "mcp-errors",
        text: "MCP Error in #context#: #exception.message#"
    )

    // Context-specific handling
    if ( context == "handleRequest" ) {
        // Additional fields available: method, requestId, params, responseTime, errorCode
        writeLog(
            type: "error",
            file: "mcp-errors",
            text: "Request failed - Method: #interceptData.method#, Error: #exception.message#"
        )

        // Send alert
        emailService.sendAlert(
            subject: "MCP Server Error",
            body: "Method: #interceptData.method#\nError: #exception.message#\nStacktrace: #exception.stackTrace#"
        )
    }
}

Registering Event Listeners

For BoxLang Module Registration

In your module's ModuleConfig.bx:

function configure() {
    interceptors = [
        {
            class: "MyMCPInterceptor",
            properties: {}
        }
    ];
}

For Application/Script Registration

Use BoxRegisterInterceptor() to register listeners:

// Application.bx
class {

    function onApplicationStart() {
        // Register self as interceptor for MCP events
        BoxRegisterInterceptor( this, "onMCPRequest,onMCPResponse,onMCPError" )

        // Create server
        MCPServer( "myApp" )
            .registerTool( myTool )
    }

    function onMCPRequest( event, interceptData ) {
        // Track incoming requests
        metrics.increment( "mcp.requests" )
    }

    function onMCPResponse( event, interceptData ) {
        // Track response times
        metrics.gauge( "mcp.responseTime", interceptData.responseTime )
    }

    function onMCPError( event, interceptData ) {
        // Handle errors
        errorTracker.captureException( interceptData.exception )
    }

}

Event Use Cases

Custom Logging

function onMCPRequest( event, interceptData ) {
    writeLog(
        type: "information",
        file: "mcp-audit",
        text: "User: #session.user.id#, Method: #interceptData.requestData.method#"
    )
}

Metrics & Monitoring

function onMCPResponse( event, interceptData ) {
    // Send to monitoring service
    metrics.gauge( "mcp.responseTime", interceptData.responseTime )
    metrics.increment( "mcp.requests", { method: interceptData.requestData.method } )
}

function onMCPError( event, interceptData ) {
    metrics.increment( "mcp.errors", { context: interceptData.context } )
}

Rate Limiting

function onMCPRequest( event, interceptData ) {
    var clientId = request.getHeader( "X-Client-ID" )

    if ( !rateLimiter.allow( clientId ) ) {
        // Reject request
        interceptData.reject = true
        interceptData.errorMessage = "Rate limit exceeded"
    }
}

Error Alerting

function onMCPError( event, interceptData ) {
    // Critical error alerting
    if ( interceptData.context == "handleRequest" ) {
        slackService.sendMessage(
            channel: "##alerts",
            text: "MCP Error: #interceptData.exception.message#"
        )
    }
}

Statistics & Monitoring 📊

The MCP server automatically tracks performance and usage metrics for real-time monitoring and analytics.

Enabling Statistics

Statistics are enabled by default when creating an MCP server:

// Stats enabled by default
server = MCPServer( "myApp" )

// Explicitly control stats tracking
server = MCPServer(
    name: "myApp",
    statsEnabled: true  // or false to disable
)

Retrieving Statistics

Get Summary Statistics

Quick overview of key metrics (lightweight):

summary = server.getStatsSummary()

writeOutput( "Total Requests: #summary.totalRequests#" )
writeOutput( "Success Rate: #summary.successRate#%" )
writeOutput( "Avg Response Time: #summary.avgResponseTime#ms" )
writeOutput( "Tool Calls: #summary.totalToolInvocations#" )
writeOutput( "Resource Reads: #summary.totalResourceReads#" )
writeOutput( "Prompts Generated: #summary.totalPromptGenerations#" )
writeOutput( "Errors: #summary.totalErrors#" )
writeOutput( "Uptime: #summary.uptime / 1000#s" )

Summary includes:

uptime - Server uptime in milliseconds
totalRequests - Total requests processed
successRate - Success rate as percentage
avgResponseTime - Average response time in milliseconds
totalToolInvocations - Total tool calls
totalResourceReads - Total resource reads
totalPromptGenerations - Total prompt generations
totalErrors - Total errors encountered
lastRequestAt - Timestamp of last request (empty if no requests)

Get Detailed Statistics

Complete stats with breakdowns:

stats = server.getStats()

// Request breakdown
writeOutput( "Successful: #stats.requests.successful#" )
writeOutput( "Failed: #stats.requests.failed#" )
writeOutput( "By Method: #serializeJSON( stats.requests.byMethod )#" )
writeOutput( "Min/Max Response Time: #stats.requests.minResponseTime# / #stats.requests.maxResponseTime#ms" )

// Tool breakdown
writeOutput( "Tools by name: #serializeJSON( stats.tools.byTool )#" )
writeOutput( "Avg Execution Time: #stats.tools.avgExecutionTime#ms" )

// Resource breakdown
writeOutput( "Reads by URI: #serializeJSON( stats.resources.byUri )#" )

// Prompt breakdown
writeOutput( "Generations by name: #serializeJSON( stats.prompts.byName )#" )

// Error breakdown
writeOutput( "Errors by code: #serializeJSON( stats.errors.byCode )#" )
if ( !stats.errors.lastError.isEmpty() ) {
    writeOutput( "Last Error: #stats.errors.lastError.message# at #stats.errors.lastError.timestamp#" )
}

Detailed stats include:

Requests: total, successful, failed, byMethod (map), response times (array), avg/min/max times, lastRequestAt
Tools: totalInvocations, byTool (map with count/totalTime/avgTime per tool), execution times (array), avg/min/max times
Resources: totalReads, byUri (map of read counts)
Prompts: totalGenerations, byName (map of generation counts)
Errors: total, byCode (map of error counts), lastError (code/message/timestamp)

Managing Statistics

Check if Stats are Enabled

if ( server.isStatsEnabled() ) {
    writeOutput( "Stats tracking is active" )
}

Enable/Disable Tracking

// Disable stats tracking (stops recording new data)
server.disableStats()

// Re-enable stats tracking
server.enableStats()

// Fluent chaining works
server.enableStats()
    .registerTool( myTool )
    .registerResource( myResource )

Reset Statistics

Clear all statistics back to zero:

// Reset all counters and data
server.resetStats()

writeOutput( "Statistics have been reset" )

// Fluent chaining
server.resetStats()
    .enableStats()

Dashboard Example

Create a real-time monitoring dashboard:

// Get server stats
server = MCPServer( "myApp" )
stats = server.getStatsSummary()

writeOutput( "
    <div class='stats-dashboard'>
        <h2>MCP Server: myApp</h2>

        <div class='stat-box'>
            <label>Uptime</label>
            <value>#numberFormat( stats.uptime / 1000, '0' )#s</value>
        </div>

        <div class='stat-box'>
            <label>Total Requests</label>
            <value>#stats.totalRequests#</value>
        </div>

        <div class='stat-box'>
            <label>Success Rate</label>
            <value>#numberFormat( stats.successRate, '0.00' )#%</value>
        </div>

        <div class='stat-box'>
            <label>Avg Response</label>
            <value>#numberFormat( stats.avgResponseTime, '0' )#ms</value>
        </div>

        <div class='stat-box'>
            <label>Tool Calls</label>
            <value>#stats.totalToolInvocations#</value>
        </div>

        <div class='stat-box error'>
            <label>Errors</label>
            <value>#stats.totalErrors#</value>
        </div>
    </div>
" )

Using Stats with Events

Combine statistics with event interception for custom monitoring:

// Application.bx
class {

    function onApplicationStart() {
        MCPServer( "myApp" )
            .registerTool( myTool )

        // Register interceptor for custom metrics
        BoxRegisterInterceptor( this, "onMCPRequest,onMCPResponse" )
    }

    function onMCPRequest( event, interceptData ) {
        // Log incoming requests
        writeLog(
            type: "information",
            file: "mcp-requests",
            text: "Request: #interceptData.requestData.method#"
        )
    }

    function onMCPResponse( event, interceptData ) {
        // Get current stats after each request
        var server = interceptData.server
        var summary = server.getStatsSummary()

        // Alert on high error rate
        if ( summary.successRate < 90 && summary.totalRequests > 10 ) {
            writeLog(
                type: "error",
                file: "mcp-errors",
                text: "High error rate: #summary.successRate#%"
            )
        }

        // Alert on slow responses
        if ( summary.avgResponseTime > 1000 ) {
            writeLog(
                type: "warning",
                file: "mcp-performance",
                text: "Slow avg response: #summary.avgResponseTime#ms"
            )
        }
    }

    function onMCPError( event, interceptData ) {
        // Handle MCP server errors
        var exception = interceptData.exception
        var context = interceptData.context

        // Log detailed error information
        writeLog(
            type: "error",
            file: "mcp-errors",
            text: "MCP Error in #context#: #exception.message# - #exception.detail#"
        )

        // Send alert for critical errors
        if ( context == "handleRequest" ) {
            // Alert operations team
            emailService.sendAlert(
                subject: "MCP Server Error",
                body: "Method: #interceptData.method#, Error: #exception.message#"
            )
        }
    }

}

Performance Notes

Zero Overhead When Disabled: When statsEnabled: false, no performance impact
Memory Efficient: Only last 1000 timing samples retained per metric
Thread Safe: Uses atomic operations for concurrent request handling
Real-Time: Stats updated immediately on each operation
Lightweight Summary: getStatsSummary() is optimized for frequent polling

API Integration Example

Expose stats via REST API:

// In your API handler
class {

    function getServerStats( required string serverName ) {
        var server = MCPServer( arguments.serverName )

        if ( isNull( server ) ) {
            return {
                success: false,
                error: "Server not found"
            }
        }

        return {
            success: true,
            data: server.getStatsSummary()
        }
    }

    function resetServerStats( required string serverName ) {
        var server = MCPServer( arguments.serverName )

        if ( isNull( server ) ) {
            return { success: false }
        }

        server.resetStats()

        return {
            success: true,
            message: "Statistics reset successfully"
        }
    }

}

Best Practices

1. Use Descriptive Names

// Good
aiTool( "searchProducts", "Search the product catalog", handler )

// Bad
aiTool( "sp", "search", handler )

2. Register at Application Start

// Application.bx onApplicationStart
// Ensures tools are available for all requests

3. Clean Up on Shutdown

// Application.bx onApplicationEnd
bxModules.bxai.models.mcp.MCPServer::removeInstance( "myApp" )

4. Use Separate Servers for Different Purposes

// API server for external clients
MCPServer( "api" )
    .registerTool( publicTool1 )
    .registerTool( publicTool2 )

// Admin server for internal tools
MCPServer( "admin" )
    .withBasicAuth( "admin", getEnv( "ADMIN_PASSWORD" ) )
    .registerTool( adminTool1 )
    .registerTool( adminTool2 )

5. Secure Sensitive Servers

// Protect production servers with authentication
MCPServer( "production" )
    .withBasicAuth( getEnv( "MCP_USER" ), getEnv( "MCP_PASS" ) )
    .setCors( "https://trusted-domain.com" )
    .registerTool( sensitiveDataTool )

// Public servers can remain open
MCPServer( "public" )
    .setCors( "*" )
    .registerTool( publicSearchTool )

Security Checklist:

✅ Use .withBasicAuth() for sensitive operations
✅ Configure CORS with .setCors() to restrict origins
✅ Store credentials in environment variables
✅ Use HTTPS in production
✅ Monitor authentication failures via onMCPError events
✅ Implement rate limiting if needed

6. Document Your Tools

aiTool( "calculateShipping", "Calculate shipping cost based on weight and destination", handler )
    .describeArg( "weight", "Package weight in pounds" )
    .describeArg( "destination", "Destination zip code" )
    .describeArg( "expedited", "Whether to use expedited shipping (true/false)" )

7. Monitor Performance with Statistics

// Enable stats for production monitoring
server = MCPServer(
    name: "production",
    statsEnabled: true
)

// Periodically check performance
scheduled task: "checkMCPPerformance", interval: "5m" {
    var stats = server.getStatsSummary()

    if ( stats.avgResponseTime > 500 ) {
        // Alert: slow responses
        notificationService.alert( "MCP server responding slowly" )
    }

    if ( stats.successRate < 95 ) {
        // Alert: high error rate
        notificationService.alert( "MCP server error rate elevated" )
    }
}

MCP Client - Consuming MCP servers
AI Tools - Creating tools
AI Agents - Using agents with tools

External Resources

PreviousRAG (Retrieval-Augmented Generation)NextMCP Client - Model Context Protocol

Last updated 13 days ago

Good afternoon

hashtag📋 Table of Contents

hashtag🎯 What is an MCP Server?

hashtag🏗️ MCP Architecture

hashtag🚀 Transport Types

hashtag🔄 Transport Flow Comparison

hashtag🌐 HTTP Transport (Web)

hashtag🖥️ STDIO Transport (Command-Line)

hashtag🎯 Choosing a Transport

hashtagQuick Start

hashtag1. Register Tools

hashtag2. Access the MCP Endpoint

hashtagCustom URL Entry Points

hashtagServer Configuration

hashtagCreating a Server

hashtagConfigure Description and Version

hashtagGet Server Info

hashtagBasic Authentication 🔒

hashtagCORS Configuration 🌐

hashtagRequest Body Size Limits 📏

hashtagCustom API Key Validation 🔑

hashtagSecurity Headers 🛡️

hashtagTool Registration

hashtagRegister a Single Tool

hashtagRegister Multiple Tools

hashtagCheck and Retrieve Tools

hashtagList Tools (MCP Format)

hashtagUnregister Tools

hashtagAnnotation-Based Discovery

hashtagScan for Annotations

hashtag@mcpTool Annotation

hashtag@mcpResource Annotation

hashtag@mcpPrompt Annotation

hashtagResource Registration

hashtagRegister a Resource

hashtagRegister Dynamic Resources

hashtagList and Read Resources

hashtagManage Resources

hashtagPrompt Registration

hashtagRegister a Prompt

hashtagList and Get Prompts

hashtagHandling MCP Requests

hashtagDirect Request Handling

hashtagHandle JSON String

hashtagSupported Methods

hashtagHTTP Endpoint (mcp.bxm)

hashtagDiscovery (GET)

hashtagJSON-RPC Requests (POST)

hashtagMultiple Servers

hashtagCORS Support

hashtagStatic Server Management

hashtagCheck Server Existence

hashtagGet All Server Names

hashtagRemove a Server

hashtagClear All Servers

hashtagComplete Example

hashtagApplication Setup

hashtagConnecting from AI Client

hashtagEvents & Interception 🎯

hashtagAvailable Events

hashtagonMCPServerCreate

hashtagonMCPServerRemove

hashtagonMCPRequest

hashtagonMCPResponse

hashtagonMCPError ⚠️

hashtagRegistering Event Listeners

hashtagFor BoxLang Module Registration

hashtagFor Application/Script Registration

hashtagEvent Use Cases

hashtagCustom Logging

hashtagMetrics & Monitoring

hashtagRate Limiting

hashtagError Alerting

hashtagStatistics & Monitoring 📊

hashtagEnabling Statistics

hashtagRetrieving Statistics

hashtagGet Summary Statistics

hashtagGet Detailed Statistics

hashtagManaging Statistics

hashtagCheck if Stats are Enabled

📋 Table of Contents

🎯 What is an MCP Server?

🏗️ MCP Architecture

🚀 Transport Types

🔄 Transport Flow Comparison

🌐 HTTP Transport (Web)

🖥️ STDIO Transport (Command-Line)

🎯 Choosing a Transport

Quick Start

1. Register Tools

2. Access the MCP Endpoint

Custom URL Entry Points

Server Configuration

Creating a Server

Configure Description and Version

Get Server Info

Basic Authentication 🔒

CORS Configuration 🌐

Request Body Size Limits 📏

Custom API Key Validation 🔑

Security Headers 🛡️

Tool Registration

Register a Single Tool

Register Multiple Tools

Check and Retrieve Tools

List Tools (MCP Format)

Unregister Tools

Annotation-Based Discovery

Scan for Annotations

@mcpTool Annotation

@mcpResource Annotation

@mcpPrompt Annotation

Resource Registration

Register a Resource

Register Dynamic Resources

List and Read Resources

Manage Resources

Prompt Registration

Register a Prompt

List and Get Prompts

Handling MCP Requests

Direct Request Handling

Handle JSON String

Supported Methods

HTTP Endpoint (mcp.bxm)

Discovery (GET)

JSON-RPC Requests (POST)

Multiple Servers

CORS Support

Static Server Management

Check Server Existence

Get All Server Names

Remove a Server

Clear All Servers

Complete Example

Application Setup

Connecting from AI Client

Events & Interception 🎯

Available Events

`onMCPServerCreate`

`onMCPServerRemove`

`onMCPRequest`

`onMCPResponse`

`onMCPError` ⚠️

Registering Event Listeners

For BoxLang Module Registration

For Application/Script Registration

Event Use Cases

Custom Logging

Metrics & Monitoring

Rate Limiting

Error Alerting

Statistics & Monitoring 📊

Enabling Statistics

Retrieving Statistics

Get Summary Statistics

Get Detailed Statistics

Managing Statistics

Check if Stats are Enabled

Enable/Disable Tracking