Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Course Outline
The Protocol Anatomy
- Why function calling alone is insufficient for complex agent ecosystems.
- MCP primitives: tools, resources, prompts, and their JSON schemas.
- Lifecycle of an MCP session: initialize, list tools, call, return, shutdown.
- Comparing MCP to OpenAPI and GraphQL for exposing capabilities to agents.
Building a Stdio MCP Server
- Scaffolding a TypeScript MCP server with the official SDK.
- Defining tool schemas with Zod and generating runtime validation.
- Implementing tool handlers that call internal REST APIs or databases.
- Handling errors, partial results, and long-running tool execution.
Building an HTTP MCP Server
- Upgrading from stdio to HTTP for remote deployment and load balancing.
- Implementing authentication using bearer tokens and mTLS.
- Graceful degradation when HTTP connections fail mid-session.
- Deploying HTTP MCP servers behind Kong or nginx with rate limiting.
Client Integration Patterns
- Registering an MCP server with Claude Code using the configuration file.
- Connecting OpenClaude to multiple MCP endpoints simultaneously.
- Writing a custom Python agent client using the MCP Python SDK.
- Gracefully handling tool availability changes at runtime.
Resource and Prompt Exposure
- Exposing read-only resources for agent context enrichment.
- Creating parameterized prompt templates that guide agent reasoning.
- Updating resources dynamically when underlying data changes.
- Separating mutable tools from immutable resources for security clarity.
Internal Tool Registry and Discovery
- Building a company-wide MCP registry with metadata and ownership tags.
- Auto-discovery via DNS-SD or well-known endpoint files.
- Versioning tools and deprecating old endpoints without breaking clients.
- Cataloging tools with natural language descriptions for agent searchability.
Enterprise Security Boundaries
- Implementing authorization checks inside tool handlers based on agent identity.
- Using network segmentation to isolate high-risk tools from general agent access.
- Sandboxing tool execution with seccomp and gVisor containers.
- Logging every tool invocation for compliance and forensic analysis.
Performance and Reliability Engineering
- Setting timeout policies per tool family: database, compute, and external APIs.
- Implementing circuit breakers when downstream services are unhealthy.
- Caching tool results to reduce redundant expensive computations.
- Running MCP servers as sidecars versus standalone microservices.
Interoperability Across Agent Platforms
- Testing MCP server compatibility with Claude Code and Continue.dev clients.
- Handling transport negotiation differences between platforms.
- Writing polyfill adapters for non-MCP agent frameworks.
- Building a cross-platform tool marketplace inside the organization.
Evolving the MCP Ecosystem Internally
- Collecting developer feedback on tool usefulness and accuracy.
- Running quarterly tool audits and pruning obsolete integrations.
- Onboarding new teams with self-service MCP server templates.
- Contributing improvements upstream to the open-source MCP specification.
Requirements
- Programming experience in TypeScript or Python.
- Understanding of LLM tool calling and function-calling patterns.
- Foundational networking knowledge: HTTP, WebSockets, and JSON-RPC.
Audience
- Backend developers constructing custom tools for AI agents.
- Platform engineers standardizing AI agent access to enterprise systems.
- Solution architects designing AI tool ecosystems for corporate adoption.
14 Hours
