Model Context Protocol

MCP

An open-source standard for connecting AI applications to external systems

MCP Specification

A specification of MCP that outlines the implementation requirements for clients and servers.

MCP SDKs

SDKs for different programming languages that implement MCP.

MCP Development Tools

Tools for developing MCP servers and clients, including the MCP Inspector.

What kind of architecture does MCP follow?

Client-server architecture

How does an AI application establish a connection to an MCP server?

By creating one MCP client for each MCP server. Each MCP client maintains a dedicated connection with its corresponding MCP server.

MCP Host

The AI application that coordinates and manages one or multiple MCP clients.

Two MCP layers

1. Data layer

2. Transport layer

Data layer

Defines the JSON-RPC based protocol for client-server communication.

Transport layer

Defines the communication mechanisms and channels that enable data exchange between clients and servers.

Discovery (Data layer)

Lets clients query a server’s supported protocol versions, capabilities, and identity through the server/discover request

Server features (Data layer)

Enables servers to provide core functionality including tools for AI actions, resources for context data, and prompts for interaction templates from and to the client.

Client features (Data layer)

Enables servers to elicit input from the user.

Utility features (Data layer)

Supports additional capabilities like notifications for real-time updates and progress tracking for long-running operations.

Stdio transport (Transport layer)

Uses standard input/output streams for direct process communication between local processes on the same machine.

Streamable HTTP transport (Transport layer)

Uses HTTP POST for client-to-server messages with optional Server-Sent Events for streaming capabilities. This transport enables remote server communication and supports standard HTTP authentication methods.

Data layer protocol

MCP uses JSON-RPC 2.0 as its underlying RPC protocol. Client and servers send requests to each other and respond accordingly.

True or false: MCP is a stateless protocol

True

Primitives

Define what clients and servers can offer each other. Specify the types of contextual information that can be shared with AI applications and the range of actions that can be performed.

Three core primitives that servers can expose

1. Tools

2. Resources

3. Prompts

Tools

Executable functions that AI applications can invoke to perform actions (e.g., file operations, API calls, database queries)

Resourrces

Data sources that provide contextual information to AI applications (e.g., file contents, database records, API responses)

Prompts

Reusable templates that help structure interactions with language models (e.g., system prompts, few-shot examples)

Elicitation

Allows servers to request additional information from users.

Notifications

MCP supports real-time notifications to enable dynamic updates between servers and clients.

Discovery

A method of allowing clients to learn what a server supports before issuing other requests.

Protocol Version Selection (Discovery)

Declares the version the client is speaking to on this request. If the server does not support the requested version, it rejects the request, lists the versions it does support, and the client retries with a mutually supported version.

Capability Discovery (Discovery)

The client declares its capabilities on every request, and the server returns its own capabilities object from server/discover. This tells each party which primitives the other can handle, and whether change notifications are available.

Identity Exchange (Discovery)

The clientInfo field in the request’s _meta and the serverInfo field in the result’s _meta provide identification and versioning information for debugging and compatibility.

Tools

Functions that your LLM can actively call, and decides when to use them based on user requests. Tool can write to databases, call external APIs, modify files, and more.

Who controls tools

The LLM model

Tool examples

Search flights, send messages, create calendar events

Resources

Passive data sources that provide read-only access to information for context, such as file contents, database schemas, or API documentation.

Resource examples

Retrieve documents, access knowledge bases, read calendars

Who controls resources

The application

Prompts

Pre-built instruction templates that tell the model to work with specific tools and resources.

Prompt examples

Plan a vacation, summarize my meetings, draft an email

Who controls prompts

The user

How tools work

Schema-defined interfaces that LLMs can invoke. They perform a single operation with clearly defined inputs and outputs.

MCP client

Instantiated by host applications to handle one direct communication with one server. Protocol-level component that enables server connections.

Two elicitation modes

Form mode, URL mode

Form mode

The server asks the client to collect structured data from the user. The request includes a schema that the client uses to build an input form and validate the response.

URL mode

The server provides a URL for the user to open. The interaction happens out of band and its data never passes through the client, making it more suitable or sensitive flows such as credential entry.

Agent Skills

Lightweight, open format for extending AI agent capabilities with specialized knowledge and workflows.

Why agent skills?

Agents often don’t have the context they need to do real work reliably. Skills solve this by packaging procedural knowledge and company, team, and user specific context into portable, version-controlled folders that agents load on demand.

Three things skills give agents

1. Domain expertise

2. Repeatable workflows

3. Cross-product reuse

Three steps to agent skills

1. Discovery

2. Activation

3. Execution