MCP implementation in practice

Your clients are asking to use ChatGPT or Claude with your data. Anthropic's MCP makes this integration possible, but implementation raises specific technical challenges.

MCP Protocol

Building an MCP server involves handling JSON-RPC messages, schema validation, authentication, and error management. Initial architecture mistakes are costly to fix.

Technical Expertise

MCP has only existed for a few months. Few developers master this technology. Training a team or recruiting this expertise takes time.

Evolving Ecosystem

Claude Desktop supports MCP natively. Other tools are gradually integrating it. Companies that implement MCP quickly gain faster access to advanced AI capabilities.

Our MCP Expertise

Our years of experience in interoperability naturally led us to MCP. Rather than reinventing the same patterns for each client, we developed mcpresso - a library that solves recurring MCP implementation problems.

mcpresso Architecture

Resource Definition

Resource DefinitionTYPESCRIPT

const noteResource = createResource({
  name: "note",
  schema: NoteSchema,
  uri_template: "notes/{id}",
  
  // Relations between resources
  relations: {
    authorId: { type: 'user' },
    tagIds: { type: 'tag' }
  },
  
  // Fine-grained operation control
  methodConfig: {
    create: { omit: ['id', 'createdAt'] },
    update: { pick: ['content', 'tagIds'] },
    search: {
      schema: z.object({
        query: z.string(),
        authorId: z.string().optional()
      })
    }
  }
});

Server Configuration

Server ConfigurationTYPESCRIPT

const server = createMCPServer({
  name: "my_server",
  resources: [noteResource],
  
  // Built-in OAuth 2.1
  auth: {
    issuer: 'https://auth.company.com'
  },
  
  // Rate limiting
  rateLimit: {
    windowMs: 15 * 60 * 1000,
    limit: 100
  },
  
  // Automatic retry
  retry: {
    retries: 5,
    factor: 2,
    maxTimeout: 60000
  },
  
  // Type exposure for AI models
  exposeTypes: true
});

Custom Methods

Custom Business MethodsTYPESCRIPT

// Custom methods for your business domain
customMethods: {
  archive_by_date: {
    description: "Archive notes older than specified days",
    inputSchema: z.object({
      days: z.number().min(1).describe("Number of days")
    }),
    handler: async ({ days }) => {
      const cutoffDate = new Date(Date.now() - days * 24 * 60 * 60 * 1000);
      return await archiveNotesOlderThan(cutoffDate);
    }
  },
  
  generate_summary: {
    description: "Generate AI summary of user's notes",
    inputSchema: z.object({
      userId: z.string(),
      period: z.enum(['week', 'month'])
    }),
    handler: async ({ userId, period }) => generateSummary(userId, period)
  }
}

This architecture reflects our understanding of production MCP challenges: OAuth 2.1 security, robustness with retry and rate limiting, extensibility with custom methods, and AI compatibility with type exposure. mcpresso allows us to implement MCP quickly while avoiding common pitfalls.

📦 mcpresso on GitHub 💬 Discuss your MCP project

MCP Implementation

We implement MCP according to your needs: compatibility analysis, pilot server development, then complete extension.

Technical Analysis

€900

1 day

Analysis of your existing APIs and evaluation of MCP implementation on your technical stack.

API compatibility with MCP
Resources exposable to AI agents
Development estimation
Recommended technical architecture

Recommended

Pilot MCP Server

€2,200

2 days

Development of a functional MCP server with mcpresso on a defined scope of your data.

mcpresso server configuration
Connection to your APIs
Testing with Claude Desktop
Technical documentation
Team training

Full Extension

€1,100/day

Variable by scope

Extension to all your resources with security and production deployment.

Implementation of all resources
Authentication and security
Monitoring and logging
Integration testing
User documentation

Your clients want to use ChatGPT with your data ?