Building MCP server

Click Here to Register

Main Speaker


Leonid Pevzner

Learning Tracks

Course ID

42892

Date

30-06-2026

Time

Daily seminar
9:00-16:30

Location

John Bryce ECO Tower, Homa Umigdal 29 Tel-Aviv
לרישום לסמינר לחצו כאן

Overview

 By the end of this seminar, participants will be able to:
  • Explain what an LLM is, what an agent is and why agents need tools/servers
  • Explain MCP (what it solves, core concepts)
  • Connect a client to an existing MCP server/workshop and run tool calls
  • Design + implement a custom MCP server with 2 tools + 1 resources
  • Add safety, validation, logging and tests

Who Should Attend

Prerequisites

  • Basic Python or TypeScript/Node.js (functions, HTTP, JSON)
  • Comfortable with CLI + env vars
  • Basic REST concepts

Course Contents

What is an LLM?
  • LLMs as probabilistic text engines + structured outputs
  • Tokens, context window, system vs user instructions
  • Why LLM apps need tools (reliability, grounding, actions) Mini-demo: same prompt with/without tool access
What is an Agent?
  • Definition: LLM + goal + loop (plan → act → observe → reflect)
  • Agent building blocks:
    • Policy / instructions (what it tries to do)
    • Memory / state (what it keeps track of)
    • Tools (how it acts in the world)
    • Environment (APIs, files, web, DB)
  • Agent patterns:
    • Single-step tool use vs multi-step planning
    • Tool selection + tool chaining
    • Guardrails: tool allowlists, budgets, timeouts
  What is MCP?
  • MCP (Model Context Protocol): why it exists (standard tool/resource interface)
  • “Client ↔ MCP Server” model and what gets standardized
  • Core primitives:
    • Tools (actions with input/output schemas)
    • Resources (readable context/data)
    • Prompts (reusable instruction templates)
  • Transport overview: local (stdio) vs remote (network) and tradeoffs
Workshop 1: Connect to an existing MCP workshop/server
  • Setup dependencies + env vars
  • Configure client to connect to the workshop MCP server
  • Discover tools/resources
  • Invoke 2–3 tools and inspect structured outputs
  • Debug checklist: schema mismatch, connection, timeouts, permissions
Designing your own MCP server (before coding)
  • Choosing tool vs resource
  • Tool contracts: input schema, output shape, error shape
  • Operational basics: idempotency, rate limiting, timeouts, caching
  • Security basics: least privilege, secrets, allowlists
Workshop 2: Build a custom MCP server
  • Scaffold project (Python)
  • Implement Tool #1 (deterministic)
  • Implement Tool #2 (safe external call or safe local access with allowlist)
  • Implement 1 resource (read-only)
  • Add structured logging + input validation + consistent errors
Deliverable: running server + client calls to both tools + resource fetch Integration + testing + guardrails
  • Contract tests: schema, success and failure modes
  • Observability basics: logs, correlation IDs
  • Injection safety basics for tools/resources (validation, sanitization, allowlists)
Wrap-up: production notes + next steps
  • Deployment options (local, container, internal service)
  • Versioning tools (don’t break clients)
  • Performance: timeouts, caching, concurrency
  • Next steps: auth, per-tool permissions, multi-tenant servers
 

For More Information, Fill In The Details

Facebook Linkedin Youtube Instagram Envelope
Copyright © 2026
לפרטים
התקשרו
לפרטים
להרשמה
התקשרו

The conference starts in

Days
Hours
Minutes
Seconds
Registration