Talki Academy
🔌

MCP (Model Context Protocol) in Production

An intensive technical training for developers who want to integrate external tools into Claude in a standardized, reusable way. From the JSON-RPC protocol to deployment on AWS Lambda, you'll build 5 complete MCP servers using the official TypeScript SDK. Each module covers a concrete production use case with runnable code, tests, and deployment scripts you can take to your own projects.

Duration
3 days
Level
Advanced
Price
9.99 EUR/month (all courses included)
Max group
12 participants

What you will learn

+Understand MCP's JSON-RPC 2.0 architecture and its core primitives: tools, resources, and prompts
+Build and publish a complete MCP server using the official TypeScript SDK in under an hour
+Implement a filesystem RAG server with incremental indexing and semantic search via Ollama
+Build a secure PostgreSQL MCP server with read-only access and SQL query validation
+Wrap REST APIs (GitHub, Stripe, internal APIs) as MCP servers with authentication and caching
+Deploy MCP servers to Docker and AWS Lambda with Langfuse monitoring

Course program

Module 1: MCP Protocol Fundamentals and Architecture

3h
  • MCP JSON-RPC 2.0 architecture: tools, resources, prompts, and transport layers
  • MCP vs direct API calls: decision criteria, latency trade-offs, reusability benefits
  • Official TypeScript SDK setup and MCP project structure
  • Tool declaration with Zod schema validation and error handling patterns
  • First working MCP server: hello world to real use case in 30 minutes
  • Debugging with Claude Desktop and MCP Inspector

Module 2: Use Case 1: Filesystem MCP for Document RAG

4h
  • RAG architecture with MCP: incremental indexing, embeddings, SQLite storage
  • Directory watching with chokidar for real-time file change detection
  • Embedding generation with Ollama (nomic-embed-text) — zero API cost
  • Semantic search with cosine similarity and hybrid BM25 search
  • Result filtering and ranking: top-K chunks, deduplication, re-ranking
  • Workshop: MCP server that indexes a documentation base and answers developer questions

Module 3: Use Case 2: PostgreSQL MCP for Database Queries

3h30
  • Security: read-only PostgreSQL user, SQL query allowlist, row limit enforcement
  • Automatic schema discovery exposed as a typed MCP resource
  • SQL validation and result row limiting to prevent runaway queries
  • Markdown result formatting for readable Claude responses
  • Connection pooling and database error handling
  • Workshop: MCP server letting Claude query an e-commerce database in plain English

Module 4: Use Case 3: REST API MCP Wrapper

3h
  • API wrapping pattern: OAuth2 / API key authentication via environment variables
  • Pagination and rate limit handling with exponential backoff
  • Response caching with configurable TTL to reduce API call volume
  • Full example: GitHub API MCP server (listRepos, searchIssues, createIssue, getPR)
  • Testing with API mocks: Jest + nock for validation without network calls
  • Workshop: MCP wrapper for an internal REST API (Notion, Airtable, or business API)

Module 5: Use Case 4: Web Scraper MCP Server

3h30
  • Playwright for JavaScript rendering: navigation, CSS extraction, screenshots
  • Handling anti-bot measures: user agents, delays, proxy rotation
  • Structured data extraction with Claude: send raw HTML, receive typed JSON
  • Server-side rate limiting to respect third-party terms of service
  • Scraping → cleaning → storage pipeline with error handling and retries
  • Workshop: competitive pricing intelligence MCP server for e-commerce monitoring

Module 6: Use Case 5: Multi-Tool Orchestration

3h
  • Multi-MCP workflow composition: chaining scraping, database, and notification tools
  • Error propagation and timeout management across multi-tool pipelines
  • Parallel call optimization: Promise.all() vs sequential based on data dependencies
  • Intermediate result patterns: shared state between tools in a workflow
  • Full real-world case: competitive pricing analysis (scraping → PostgreSQL → Slack report)
  • Workshop: orchestrate 3 MCP servers on a business case of your choice

Module 7: Deployment: Docker, AWS Lambda, and Production Monitoring

2h30
  • Docker for always-on MCP servers: multi-stage Dockerfile, secrets, health checks
  • AWS Lambda for on-demand MCP servers: cold start < 2s, timeout and memory tuning
  • Environment variables and secrets management with AWS SSM Parameter Store
  • Production monitoring with Langfuse: trace every tool call, latency, and error
  • Alerting: error rate thresholds, cost spike detection, PagerDuty notifications
  • Workshop: deploy one of your 5 MCP servers to AWS Lambda with full observability

Ready to get started?

9.99 EUR/month — All courses included, cancel anytime

Request a quoteView all courses