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 inadequate for complex agent ecosystems.
- MCP primitives: tools, resources, prompts, and their JSON schemas.
- Lifecycle of an MCP session: initialization, listing tools, invocation, result return, and shutdown.
- Comparing MCP with OpenAPI and GraphQL for exposing capabilities to agents.
Building a Stdio MCP Server
- Scaffolding a TypeScript MCP server using the official SDK.
- Defining tool schemas with Zod and generating runtime validation.
- Implementing tool handlers that invoke internal REST APIs or databases.
- Managing errors, partial results, and long-running tool executions.
Building an HTTP MCP Server
- Transitioning from stdio to HTTP for remote deployment and load balancing.
- Implementing authentication via bearer tokens and mTLS.
- Gracefully degrading when HTTP connections fail during a 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.
- Handling changes in tool availability gracefully at runtime.
Resource and Prompt Exposure
- Exposing read-only resources for agent context enrichment.
- Creating parameterized prompt templates that guide agent reasoning.
- Dynamically updating resources when underlying data changes.
- Distinguishing between mutable tools and immutable resources for security clarity.
Internal Tool Registry and Discovery
- Building a company-wide MCP registry with metadata and ownership tags.
- Enabling auto-discovery via DNS-SD or well-known endpoint files.
- Versioning tools and deprecating old endpoints without disrupting clients.
- Cataloging tools with natural language descriptions to enhance agent searchability.
Enterprise Security Boundaries
- Implementing authorization checks within tool handlers based on agent identity.
- Utilizing network segmentation to isolate high-risk tools from general agent access.
- Sandboxing tool execution using seccomp and gVisor containers.
- Logging every tool invocation for compliance and forensic analysis.
Performance and Reliability Engineering
- Establishing timeout policies per tool family: database, compute, and external APIs.
- Implementing circuit breakers when downstream services are unhealthy.
- Caching tool results to minimize 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.
- Addressing transport negotiation differences between platforms.
- Writing polyfill adapters for non-MCP agent frameworks.
- Developing a cross-platform tool marketplace within the organization.
Evolving the MCP Ecosystem Internally
- Gathering developer feedback on tool usefulness and accuracy.
- Conducting 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
- Proficiency in programming with TypeScript or Python.
- Familiarity with LLM tool calling and function-calling patterns.
- Foundational networking knowledge: HTTP, WebSockets, and JSON-RPC.
Audience
- Backend developers creating 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
