Model Context Protocol (MCP) Design Patterns logopeterindia.net
Knowledge Hub

Model Context Protocol (MCP) Design Patterns

Six common ways to structure MCP servers — from a thin wrapper over one API to hierarchies of domain servers — with the specification, SDKs and security guidance behind them.

46 topics•6 categories•Updated October 2026

Choosing an MCP Design Pattern

3 entries

The Model Context Protocol (MCP) lets an AI agent discover and call tools, read resources and use prompts exposed by servers. How a server is structured matters as much as the protocol. Six patterns come up repeatedly. A direct API wrapper maps tools one-to-one onto an existing API and suits stable APIs and quick integrations. A composite service hides several APIs behind higher-level tools for complex workflows. An MCP-to-agent server calls specialist agents for focused reasoning. An event-driven server publishes events or runs long operations asynchronously so the agent does not block. A hierarchical setup puts a domain-level server in front of sub-domain servers. A local resource access server runs on the same machine for files and other local data. These patterns can be combined; this page links each to the specification and tools that support it. The six-pattern framing follows an infographic shared by Rakesh Gohel (@rakeshgohel01); the descriptions here are written independently.

01What is MCP?Official introduction to the Model Context Protocol. 02MCP architectureHosts, clients and servers, and how they connect. 03MCP specificationThe protocol specification, with versioned releases.

The Six Patterns

6 entries

Each pattern with a resource that supports it.

04Direct API wrapperMaps MCP tools one-to-one to an existing API. FastMCP can generate an MCP server from an OpenAPI specification. 05Composite serviceOne tool orchestrates several APIs behind a single higher-level capability. See the tools specification for how servers define and return tool results. 06MCP-to-agentA tool delegates to a specialist agent for reasoning. The Agent2Agent (A2A) protocol is an open standard for agent-to-agent communication that complements MCP. 07Event-driven integrationTools start work that finishes later, or publish events, so the agent is not blocked. The Tasks extension covers long-running operations in MCP. 08Hierarchical serversA domain-level server fronts sub-domain servers. FastMCP documents composing and mounting servers. 09Local resource accessServers run on the user's machine, typically over the stdio transport, to reach files and local tools.

Specification & Core Concepts

11 entries

The protocol features patterns are built from.

10Server conceptsTools, resources and prompts that servers expose; transports are covered under the local pattern above. 11ResourcesExposing data and content to clients. 12PromptsReusable prompt templates. 13SamplingServers asking the client's model to generate text. Deprecated in the 2026-07-28 specification: it still works, but new implementations should not add it. 14RootsTelling servers which locations they may work in. Deprecated in the 2026-07-28 specification; pass directories or files through tool parameters, resource URIs or server configuration instead. 15Specification repositorySource of the specification and schema. 16Key changes in specification 2026-07-28Stateless requests with no initialize handshake, no protocol-level sessions, Tasks moved to an extension, multi round-trip requests, and deprecation of Roots, Sampling and Logging. 17Deprecated features registryFeatures still in the specification but scheduled for removal, with a minimum twelve-month deprecation window. 18Multi round-trip requestsHow a server asks for more input by returning an input-required result that the client answers on retry; replaces server-initiated requests. 19Streamable HTTP transportThe HTTP transport for remote servers; the older HTTP+SSE transport is deprecated. 20MCP extensions overviewOptional, opt-in extensions such as Tasks, MCP Apps and Skills over MCP.

SDKs, Frameworks & Tools

10 entries

Build, test and publish MCP servers.

21Build an MCP serverOfficial tutorial for building a server. 22FastMCPPython framework for building MCP servers and clients. 23FastMCP proxyProxy one MCP server through another. 24MCP Python SDKOfficial Python SDK. 25MCP TypeScript SDKOfficial TypeScript SDK. 26MCP Java SDKOfficial Java SDK. 27MCP C# SDKOfficial C# SDK. 28MCP InspectorVisual tool for testing and debugging servers. 29Reference serversReference and community MCP servers, including a filesystem server. 30Official MCP RegistryRegistry for discovering MCP servers.

Security & Governance

4 entries

Authorization and risks that grow with each pattern.

31MCP authorizationAuthorization for HTTP-based transports. 32Security best practicesOfficial guidance on risks such as confused deputy and token passthrough. 33OWASP MCP Top 10OWASP project listing top MCP security risks. 34OWASP GenAI Security ProjectGuidance on security for generative AI and agentic systems.

Related Pages

12 entries

Neighbouring pages on this site.

35Top MCP ServersDirectory of MCP servers. 36Agentic AI Communication ProtocolsMCP, A2A and related agent protocols. 37AI Agent GatewaysGateways in front of agents and tools. 38AI Agent Security SolutionsSecuring AI agents. 39AI Agent FrameworksFrameworks for building agents. 40Agentic AIAgentic AI hub. 41EDA PatternsEvent-driven architecture patterns. 42Event-Driven MicroservicesEvent-driven microservice architectures. 43Software Architecture PatternsArchitecture patterns. 44Artificial IntelligenceAI hub. 45MCP Servers for AI DevelopersTen MCP servers developers connect to coding agents. 46Advanced Architectural Patterns for Agentic AISix patterns for structuring agent systems, with context engineering and evaluation.

Frequently Asked Questions

What are MCP design patterns?

They are recurring ways of structuring MCP servers: wrapping one API directly, composing several APIs into higher-level tools, delegating to specialist agents, handling events and long-running work asynchronously, organizing servers in a hierarchy, and exposing local resources. The names are conventions, not part of the specification.

When should I use a direct API wrapper?

When you already have a stable API and want agents to use it quickly. Mapping each endpoint to a tool is simple, but a long list of fine-grained tools can overwhelm a model, so consider a composite service if workflows span many calls.

What is the difference between MCP and A2A?

MCP connects an agent to tools and data exposed by servers, while the Agent2Agent protocol is for communication between agents. An MCP-to-agent pattern combines them by letting a tool call another agent.

How do I handle long-running work in MCP?

Use an asynchronous approach so the agent does not block, for example the Tasks extension documented by MCP, or tools that start a job and return an identifier to poll. Check the current specification for supported features.

Is a local MCP server safer than a remote one?

Not automatically. A local stdio server avoids network exposure but still runs with the user's permissions and can read or change local files, so limit its scope with least privilege and explicit paths (the Roots feature is deprecated in the current specification). Remote servers need authorization and the guidance in the MCP security best practices.

How current is this page?

Links were checked in October 2026. The pattern names below are common ways of describing how MCP servers are used, not terms defined by the MCP specification, which changes over time. Sites that block automated checks are kept and may need a manual click.