Introducing the Model Context Protocol
by Anthropic
The November 2024 announcement that defined MCP: read it for design intent, not API detail.
Overview
Published on 25 November 2024, this is Anthropic's original announcement of the Model Context Protocol, an open standard for connecting AI assistants to the systems where data actually lives: content repositories, business tools and development environments. The post frames the problem as fragmentation, where every new data source previously required its own bespoke connector, so even frontier models stayed cut off from the information around them. MCP replaces that with a single protocol, and the announcement ships three concrete things: the protocol specification together with TypeScript and Python SDKs, local MCP server support in the Claude Desktop apps, and an open-source repository of pre-built servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer. Architecturally it describes hosts, clients and servers communicating across a protocol layer and a transport layer, with servers retaining control of their own resources so API keys never have to be handed to the model provider. Early adopters named in the post include Block and Apollo, plus the developer-tooling companies Zed, Replit, Codeium and Sourcegraph; Block CTO Dhanji R. Prasanna is quoted calling open technologies like MCP 'the bridges that connect AI to real-world applications'. At launch MCP connected only to local servers, with remote servers and enterprise authentication described as still in development.
At a Glance
- Topic
- MCP
- Level
- Beginner
- Format
- Guide
- Cost
- Free
- Duration
- ~8 min read
- Provider
- Anthropic
- Hands-on
- No
- Certificate
- None
What You’ll Learn
- ✓Understand the fragmented-integration problem MCP was designed to eliminate
- ✓Name the three MCP roles, hosts, clients and servers, and how they communicate
- ✓Distinguish the protocol layer from the transport layer in MCP's architecture
- ✓Recall what shipped at launch: specification, TypeScript and Python SDKs, reference servers
- ✓Explain why servers keep their own credentials instead of sharing API keys upward
- ✓Trace which vendors adopted MCP first and what that signalled for the ecosystem
- ✓Recognise the original local-only limitation that later remote-server work removed
Highlights
- •The primary source: every later MCP explainer is downstream of this post
- •States design intent, which the reference specification deliberately does not
- •Names the six day-one reference servers, a useful map of what MCP was built for
- •Short enough to read before a design meeting, unlike the specification itself
- •Dated November 2024, so it anchors the MCP ecosystem timeline precisely
Who It’s For
Best For
- ✓Engineers about to read the MCP specification who want the why first
- ✓Technical leaders evaluating whether to standardise on MCP
- ✓Anyone writing or reviewing MCP explainers who needs the primary source
Prerequisites
- •None: the post assumes no prior MCP knowledge
- •General familiarity with LLM assistants and API integrations helps
FAQ
What is Introducing the Model Context Protocol?
Anthropic's 25 November 2024 launch post for the Model Context Protocol, the document the entire MCP ecosystem traces back to. It explains why fragmented, bespoke integrations between AI systems and data sources needed replacing with one open standard, what shipped on day one, and which companies had already adopted it, all context the reference documentation assumes you already have.
Is Introducing the Model Context Protocol free?
Introducing the Model Context Protocol is free to access.
What level is Introducing the Model Context Protocol for?
Introducing the Model Context Protocol is aimed at a beginner audience. Recommended background: None: the post assumes no prior MCP knowledge, General familiarity with LLM assistants and API integrations helps.
How long does Introducing the Model Context Protocol take?
Expect roughly ~8 min read. Most learners work through it at their own pace.
What will I learn from Introducing the Model Context Protocol?
You'll learn: Understand the fragmented-integration problem MCP was designed to eliminate; Name the three MCP roles, hosts, clients and servers, and how they communicate; Distinguish the protocol layer from the transport layer in MCP's architecture; Recall what shipped at launch: specification, TypeScript and Python SDKs, reference servers; Explain why servers keep their own credentials instead of sharing API keys upward; Trace which vendors adopted MCP first and what that signalled for the ecosystem; Recognise the original local-only limitation that later remote-server work removed.
Topics
Sources
This page was written from 2 sources, 1 on domains other than anthropic.com.