MCP Overview

Model Context Protocol (MCP) connects AI applications to external tools and data sources in a standardized way.

Analogy: Just as USB-C connects chargers, monitors, and external drives through a single port, MCP connects AI to AWS documentation, databases, monitoring tools, and more through a single standard interface.

Without MCP: Fragmented Integration

Without MCP, M AI apps need to individually integrate with N tools, requiring M×N integrations. Each app builds custom connectors for every data source.

With MCP: Standardized Integration

With MCP, each app and tool only needs to implement one standard protocol, reducing to M+N implementations instead of M×N custom integrations.

Using the USB-C hub analogy: MCP servers connect various data sources like USB-C cables, and the MCP client acts as a hub providing a unified interface to the AI host.

With MCP, the following tasks are possible:

• Access specialized knowledge bases and documentation (e.g., AWS Documentation MCP server)
• Integrate with external services and APIs (e.g., GitHub, Slack, databases)
• Extend AI capabilities with domain-specific tools (e.g., CloudWatch metrics analysis)
• Create custom tools for specific workflows (e.g., infrastructure diagnostics prompts)

For more details, see the Kiro CLI MCP documentation.

Architecture

Component	Description	In this guide
MCP Host	AI application that accesses data through MCP	Kiro CLI
MCP Client	Protocol client within the Host, maintaining 1:1 connection with server	Built into Kiro CLI
MCP Server	Lightweight program exposing tools and data via standard MCP protocol	AWS Documentation, etc.
Data Sources	Local (files, DB) and remote (API) data accessed by MCP Server	AWS docs, CloudWatch metrics, etc.

Kiro CLI’s role

Kiro CLI acts as both a client and orchestrator in the MCP ecosystem:

• MCP Server management — Reads server definitions from configuration (mcp.json or agent JSON) to start and manage servers
• Communication mediation — Mediates communication between the LLM and MCP Server
• Tool provision — Validates and delivers tools exposed by MCP Servers for the LLM to use

Configuration methods

Method	Description
Command line	kiro-cli mcp add
Configuration file	Workspace <project>/.kiro/settings/mcp.json or global ~/.kiro/settings/mcp.json
Agent configuration	Per-agent mcpServers field in ~/.kiro/agents/{name}.json

IDE vs CLI MCP configuration

Surface	Config location	Verify with
Kiro IDE	.kiro/settings/mcp.json (workspace)	Kiro panel server status
Kiro CLI global	~/.kiro/settings/mcp.json	/mcp in CLI session
Kiro CLI agent	mcpServers in agent JSON	/mcp in CLI session

For IDE workspace MCP setup, see MCP Servers.

Request Flow

1. User → asks Kiro CLI a question
2. Kiro CLI → sends question + list of available MCP tools to LLM
3. LLM → decides which MCP Tool to call
4. MCP Client → requests tool execution from MCP Server
5. MCP Server → retrieves results from data source
6. Results returned → LLM generates final response → User

Three Core MCP Server Capabilities

Capability	Description	Controlled by	Example
Resources	Server exposes data (files, DB records)	Client decides when to use	AWS documentation content
Prompts	Reusable prompt templates and workflows	User selects and runs	@cloud_diagnosis
Tools	LLM performs actions through the server	Model decides automatically	search_documentation

This guide primarily uses Tools and Prompts. In AWS MCP Integration, you will practice AWS Documentation MCP tools and an optional custom MCP Prompt server.

Security Model

Principle	Description
Explicit Permission	User approval required before tool execution
Local Execution	MCP servers run locally
Isolation	Each MCP server runs as a separate process
Transparency	Users can check available tools

mcp.json Configuration

Scope	Path
Global	~/.kiro/settings/mcp.json
Workspace	<project>/.kiro/settings/mcp.json

MCP Server Transport Types

Local server (stdio-based): Runs as a local process and communicates via standard I/O (stdin/stdout). Each agent spawns its own MCP server subprocess. This is the method used in this guide.

HTTP server (streamable HTTP): Connects to a remote endpoint over the network. Multiple clients can share one remote server. Supports GET, POST, and Server-Sent Events (SSE).

{
  "mcpServers": {
    "my-server": {
      "command": "uvx",
      "args": ["my-mcp-server@latest"],
      "env": { "API_KEY": "your-key" }
    }
  }
}

Remote HTTP servers support OAuth authentication:

{
  "mcpServers": {
    "remote-server": {
      "url": "https://endpoint.example.com",
      "type": "http",
      "headers": { "Authorization": "Bearer token" }
    }
  }
}

Lab: Check MCP Status

Inside a Kiro CLI session:

/mcp

Expected output if no servers are configured yet:

⚠ No MCP servers installed

To learn how to configure MCP servers, visit https://kiro.dev/docs/cli/mcp

Since no MCP servers have been configured yet, this message is normal. In the next step you will add the AWS Documentation MCP server to your cloud-ops agent.

Checkpoint

You have completed this step if you have confirmed all of the following:

1. Understanding of MCP’s 3-tier Host-Client-Server structure
2. Understanding of local vs remote MCP Server configuration methods
3. Current MCP server status confirmed with /mcp command

Next: AWS MCP Integration — add the AWS Documentation MCP server to your cloud-ops agent.