Skip to main content
What is MCP? MCP stands for Model Context Protocol. It’s a lightweight, open framework that standardizes how large language models (LLMs) request data and perform actions with external systems. Think of MCP as a common interface — like a universal connector — that lets models interact predictably with databases, file systems, web APIs, and other tools.
A slide titled "MCP — Introduction" showing a Model Context Protocol diagram with bidirectional arrows between an AI Model (robot and atom icon) and an External Data Source (globe and database icons). The caption describes it as a framework for enhancing interactions between AI models and external data sources or tools.
MCP was initiated by Anthropic (the creators of Claude), but it’s designed to be an open, vendor-neutral protocol usable with any LLM. A helpful analogy: MCP is like USB for AI — a predictable, standardized way to connect models to external services.
MCP enables LLMs to access fresh, private, and actionable information without hardwiring bespoke integrations into every application.
Why MCP matters Out of the box, LLMs face practical limits that matter for real applications:
  • Training cutoff: Models only “know” information up to their training cutoff date.
  • No live data access: They cannot query the web or enterprise systems without an external bridge.
  • No direct access to private sources: Internal documents and databases require a secure interface to expose data.
  • Limited actionability: Models alone don’t perform external actions (send emails, update systems, or run scripts).
Without a protocol like MCP, teams build many custom adapters, which multiplies engineering effort and increases maintenance overhead. MCP reduces duplication by offering a consistent contract between LLMs (via a host) and external services. What we’ll cover in this lesson/article
  • Key components of MCP architecture
  • How MCP enables the flow between LLMs, host applications, and external services
  • MCP primitives: resources, tools, and prompts
  • Real-world examples and practical benefits
Below is a high-level architecture and flow overview.
A dark-slide diagram titled "How MCP Works" showing two sections: "01 Key Components" listing MCP Client/Host, MCP Protocol, MCP Server, and External Services, and "02 The Flow" listing Language Model and Client Application.
Client–server model (the core idea) At its heart, MCP uses a familiar client–server pattern to mediate access to tools and data. A restaurant analogy often makes this easy to remember:
  • Customer (Host application + LLM): The user-facing app that needs data or an action performed.
  • Waiter (MCP client): The messenger that formats and forwards requests using MCP.
  • Kitchen (MCP server): The system with access to ingredients (external data and tools) that fulfills requests.
  • Ingredients / tools (External services): Databases, file stores, web APIs, and other services.
The MCP client (waiter) sends standardized requests to one or more MCP servers (kitchens). Each server interfaces with the external services (ingredients) to fetch data or execute actions and returns results to the client, which then passes them back to the host and the LLM.
A diagram titled "How Does MCP Work?" using a restaurant analogy: the left panel shows "The Restaurant" with the client and waiter, and the right panel shows "The Kitchen (The Server)" with the chef and ingredients labeled as external data/tools.
A simplified MCP flow Typical request sequence: Host application (with the LLM) → MCP client → MCP server → External data/tool
  1. The host and LLM determine additional data or an action is required.
  2. The LLM issues an MCP-formatted request to the MCP client.
  3. The MCP client routes the request to the correct MCP server.
  4. The server communicates with the external resource, retrieves data or performs the action, and returns the result.
  5. The client relays the result back to the host application and the LLM.
This separation keeps responsibilities clear and makes it easy to add new capabilities without changing the LLM or host logic.
A diagram titled "How MCP Works" showing a vertical flow: Host Application with LLM → MCP Client → MCP Server → External Data/Tool. To the right is a resources column with items like Database, File System, and Third‑Party Web App.
MCP primitives: resources, tools, and prompts MCP servers expose three fundamental primitives that LLMs use through the client. These primitives make interactions structured, auditable, and reusable.
PrimitivePurposeCommon examples
ResourcesReadable context data the LLM can queryFile contents, database rows, API responses
ToolsActions the LLM can request the server to performSend email, create task, run a script, post to a service
PromptsStandardized templates or interaction patternsPredefined request templates, structured input forms
These primitives let the LLM request information or trigger actions in a consistent way across many backends.
A diagram titled "MCP Primitives" showing an MCP Server that connects to three categories: Resources (data), Tools (actions) and Prompts (templates). The Tools column lists example actions like sending an email, creating a task, adding an event, and running a script.
Extending the analogy In the restaurant metaphor, the host doesn’t need to understand how the kitchen prepares a dish or where every ingredient is sourced. Similarly, MCP abstracts away the internal details of external systems: the LLM “orders” data or an action, and the MCP server returns a consistent, usable response. Real-world examples MCP supports practical workflows across many domains. Typical use cases include:
  1. File system access — securely read and write files on behalf of the model, with audit logs.
  2. GitHub integration — summarize commits, create pull requests, or comment on issues programmatically.
  3. Web search augmentation — incorporate recent web results that are outside the model’s training data.
  4. Database queries — fetch private customer data to provide contextually accurate answers.
A host application can connect to multiple MCP servers (e.g., one server for files, another for databases, a third for web search), enabling a single LLM-driven interface to orchestrate many backend systems.
A presentation slide titled "MCP in Action: Real-World Examples" showing four numbered cards: 01 File System Access, 02 GitHub Integration, 03 Web Search, and 04 Database Access, each with a small icon. The cards are arranged horizontally under a dark header on a light background.
Benefits of MCP MCP delivers concrete advantages when building LLM-powered applications:
BenefitWhat it enables
StandardizationReusable integrations, less duplicated effort
Separation of concernsHost and LLM focus on intent; servers handle integrations
ExtensibilityAdd new servers for more services without changing the client
Reduced hallucinationsVerify facts by pulling authoritative data before response generation
Enhanced capabilitiesCombine LLM reasoning with real-world actions and live data
A presentation slide titled "MCP – Benefits" showing five numbered dark cards across the page. The cards list: 01 Standardization, 02 Separation of concerns, 03 Extensibility, 04 Reduced hallucinations, and 05 Enhanced capabilities.
Final thoughts MCP shifts how we apply LLMs: from isolated models with static knowledge to connected agents that can securely access live data and perform actions. By standardizing the interface between models and external systems, MCP reduces engineering friction, improves reliability, and unlocks new capabilities.
MCP is an open, standard approach for connecting models to external tools and data. In this lesson/article we’ll explore how to design and implement MCP clients and servers and how to use these primitives effectively.
Links and references

Watch Video