A New API Standard for chaining AI -- Model Context Protocol

Anthropic launched the Model Context Protocol on November 25, 2024, [8cac05] it was a game-changer for AI use and code generation. It allowed for the chaining of AI operations, which made it possible to create complex workflows that could be used to generate code, documents, and other content.
Related: Model Context Protocol, Anthropic, Chain of Draft

What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard developed to enable secure, two-way integration between AI-powered applications (such as large language models and agents) and external data sources, tools, and services. Its primary goal is to standardize how AI systems access, retrieve, and act on real-world data, replacing the fragmented, custom integrations that previously dominated the ecosystem. [8cac05] [b51563] [b479a8] [081ad2]
MCP draws inspiration from the Language Server Protocol (LSP), which unified how code editors interact with programming languages. Similarly, MCP provides a universal API for connecting AI models with external systems, transforming the integration challenge from an “M×N” problem (each app to each tool) to an “M+N” problem (apps and tools each implement MCP once). [b479a8] [026f85] [081ad2]

How MCP Works

  • Architecture: MCP uses a client-server model:
  • Host applications: LLM-powered tools (e.g., Claude Desktop, Copilot, AI IDEs).
  • MCP clients: Embedded in hosts, these handle communication with MCP servers.
  • MCP servers: Expose data, tools, or functions to AI systems via a standardized interface.
  • Transport: Communication occurs over JSON-RPC 2.0, using either local (STDIO) or remote (HTTP + SSE) channels. [b51563] [026f85]
  • Core Features:
  • Resources: Read-only data sources (like files, databases, knowledge bases).
  • Tools: Functions or APIs the AI can call to perform actions.
  • Prompts: Predefined templates to optimize how tools/resources are used. [b479a8] [026f85]
  • Security & Consent: MCP emphasizes explicit user consent, robust authorization, and clear UI for data access and tool usage. [026f85] [91430e]

Who Has Adopted MCP?

MCP has seen rapid adoption across major technology companies, AI platforms, and open-source communities. Below is a summary of notable adopters and their use cases:
Company/ServiceAdoption DetailsSource/Announcement
AnthropicCreator of MCP; integrated into Claude Desktop and open-sourced core SDKs and servers [8cac05] [b9fb3b]
Block (Square)Early adopter, using MCP to build agentic systems for business automation [8cac05]
ApolloIntegrated MCP for enhanced AI-driven workflows [8cac05]
MicrosoftAdopted MCP in Copilot Studio and Azure AI Foundry Agent Service for seamless agent integration across Microsoft 365 [b778b7] [3222d7]
Amazon AWSIntegrated MCP into AWS Bedrock agents for enterprise-scale, context-aware AI [b778b7] [8d2e1f]
GitHubAdded MCP server support to GitHub AI assistants and VS Code extensions [b778b7] [cecf76]
DeepsetUses MCP to power context-aware RAG and AI pipelines [b778b7]
AtlassianAnnounced MCP support for connecting structured knowledge to AI tools [b055c1]
CloudflareIntegrated MCP for AI-driven security and automation workflows [b9fb3b]
Zed, Replit, Codeium, SourcegraphEnhanced AI coding assistants with MCP for deeper context and tool integration [8cac05] [ce07e9]
OpenAI, Google DeepMindAnnounced MCP support for their AI platforms [081ad2] [cecf76]
Windows 11Previewed MCP as a foundational layer for secure, interoperable agentic apps [91430e]

Recent Company Blog Announcements

Summary

Model Context Protocol is rapidly becoming the backbone for connecting AI models with real-world data and tools in a secure, standardized way. Its adoption by leading cloud, productivity, and development platforms signals a shift toward more interoperable, agentic AI systems across the industry. [8cac05] [b778b7] [081ad2]

Footnotes