liveBook · Manning

MCP host

As explained in Designing AI Agents

An MCP host is an agent runtime component that connects an agent to Model Context Protocol capabilities and manages their use.

An MCP host is the runtime’s standardized interface to external tools. It supports capability discovery through JSON schemas, tool invocation, and structured results, while keeping tool implementations abstract. section 2.2.2

Table 2.3 The LLM-as-OS analogy
Computer OS component
Agent runtime equivalent
Purpose
Process isolation/sandbox
Execution sandbox
Contain the blast radius of agent actions
File system
State persistence
Survive crashes, resume tasks
IPC/message bus
Queue/event system
Buffer async tool results
System monitor/metrics
Observability layer
Track agent behavior, latency, and cost
Device drivers
MCP host
Standardized interface to external tools
Package manager
Agent skills
Discoverable, composable capability modules

The operating-system analogy maps device drivers to the MCP host, framing it as a standardized interface to external tools. section 2.2.2

How Designing AI Agents covers it

How deep it goes
Defines JSON-schema capability discovery, standardized invocation, and structured results while abstracting tool implementations.
Examples
Connects to file systems, databases, APIs, browsers, and other agents through the same protocol.
What you can do after
Apply MCP to discover capabilities, invoke tools, and receive structured results.
How it connects
Pairs with Agent Skills in the runtime VM and contributes to the agent harness.

Also explained in

Compare how 2 books explain MCP host →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Designing AI Agents. The section links point to the parts of the book it draws on.