Ch 1 — What Is MCP & Why It Exists

The integration problem, the USB-C analogy, and how one protocol changes everything
IndexUnder the Hood →
High Level
-
Click play or press Space to begin...
Step- / 8
AThe N × M Integration ProblemEvery app needs a custom connector for every tool
1
smart_toy
AI App AClaude Desktop,
Cursor, ChatGPT...
custom
storage
Tool 1GitHub API
custom wrapper
custom
database
Tool 2Postgres DB
custom wrapper
custom
cloud
Tool 3Slack, Jira...
custom wrapper
2
arrow_downward What if there was one standard protocol?
BMCP: One Protocol to Connect Them AllThe “USB-C for AI” — standardized tool integration
smart_toy
Any AI AppSpeaks MCP
once
MCP
hub
MCP ProtocolStandard wire format
JSON-RPC 2.0
MCP
extension
Any MCP ServerGitHub, Postgres,
Slack, Jira...
3
arrow_downward Inspired by a proven pattern: Language Server Protocol
CThe LSP AnalogyLSP solved editor × language — MCP solves AI app × tool
edit_note
LSPN editors × M languages
→ N + M adapters
same idea
hub
MCPN AI apps × M tools
→ N + M adapters
result
4
trending_down
N + M, Not N × MBuild once, connect
to everything
5
arrow_downward What can an MCP server expose?
DThe Three PrimitivesTools, Resources, and Prompts
build
ToolsFunctions the model
can call (actions)
+
description
ResourcesData the model
can read (context)
+
chat
PromptsReusable templates
and workflows
6
arrow_downward Who are the players in the protocol?
EThe Three RolesHost, Client, and Server
laptop
HostThe AI application
(Claude, Cursor)
manages
swap_horiz
ClientConnector inside host
1 per server
talks to
dns
ServerExposes tools,
resources, prompts
7
arrow_downward Who adopted MCP and where is it going?
FAdoption & the Road AheadFrom Anthropic to the Linux Foundation
rocket_launch
Nov 2024Anthropic open-sources
MCP specification
→
groups
Mass AdoptionOpenAI, Google, MS,
Cursor, Replit...
→
8
foundation
Linux FoundationDonated to Agentic
AI Foundation
1
Detail