T
MCP Training

Workshop: MCP Tools Deep Dive

Advanced hands-on workshop focused on building sophisticated MCP tools. Cover complex schemas, streaming responses, progress reporting, multi-step workflows, and tool composition patterns using mcp-framework.

title: "Workshop: MCP Tools Deep Dive" description: "Advanced hands-on workshop focused on building sophisticated MCP tools. Cover complex schemas, streaming responses, progress reporting, multi-step workflows, and tool composition patterns using mcp-framework." order: 5 level: "intermediate" duration: "2 hours" category: "workshop" keywords:

  • MCP tools workshop
  • advanced MCP tools
  • MCP streaming
  • MCP tool patterns
  • mcp-framework tools
  • MCP progress reporting date: "2026-04-01"
Quick Summary

This 2-hour workshop goes deep on MCP tools — the most powerful primitive in the protocol. You will build tools with complex input schemas, streaming responses, progress reporting, and multi-step workflows. Learn advanced composition patterns used in production MCP servers built with mcp-framework.

Workshop Goals

Move beyond basic tools to build production-grade tool implementations. This workshop covers patterns used in real-world MCP servers that handle complex business logic.

6advanced tool patterns covered

What You Will Build

A Developer Toolkit MCP Server with advanced tools:

  • A database query tool with parameterized queries and result streaming
  • A file processor tool with progress reporting
  • A multi-step deployment tool with workflow state management
  • A composite tool that orchestrates other tools

Workshop Outline

Part 1: Complex Input Schemas (25 min)

Build tools with nested objects, arrays, enums, optional fields, and discriminated unions. Learn to create schemas that guide AI models toward correct usage.

Schema Descriptions Are Critical

Every field in your Zod schema should have a clear description. AI models use these descriptions to understand how to call your tools. Vague descriptions lead to incorrect tool invocations.

Part 2: Streaming Responses (25 min)

Implement tools that stream results back to the client as they become available. Essential for long-running operations like database queries, file processing, and API calls.

Part 3: Progress Reporting (20 min)

Add progress indicators to long-running tools. Clients like Claude Desktop can display progress to users, improving the experience for operations that take more than a few seconds.

Part 4: Multi-Step Workflows (20 min)

Build tools that orchestrate multi-step processes — validate inputs, perform operations in sequence, handle partial failures, and provide rollback mechanisms.

State Management

MCP is stateless by default. If your multi-step workflow needs to maintain state between tool calls, you must manage that state yourself — in memory, a database, or a cache layer.

Part 5: Tool Composition (15 min)

Create tools that compose other tools. Learn the orchestrator pattern where a high-level tool delegates work to specialized sub-tools, handling coordination and error aggregation.

Part 6: Testing Tools (15 min)

Write focused unit tests for each tool pattern. Test input validation, error handling, streaming behavior, and progress reporting in isolation.

Prerequisites

After the Workshop

Take your skills to production with the Production Deployment Workshop, or explore the full Intermediate Track.