Model Context Protocol

• 1. Introduction

  ›

  The Model Context Protocol, usually shortened to MCP, was built to solve a problem that Anthropic summed up with a piece of arithmetic: N×M. Imagine many AI systems on one side and many tools and data sources on the other. Every pairing needed its own bridge. MCP is an open standard and open-source framework meant to replace that tangle with one shared way for language models to reach beyond themselves. It sounds like plumbing, yet it pulled rival companies into the same camp and drew sharp warnings from security researchers. Why did a problem this unglamorous matter so much? Who designed the answer? How did competitors come to adopt a standard that Anthropic created, and what did researchers find when they inspected it?

  516

• 2. The Bridge Problem

  ›

  Before MCP, a developer who wanted an AI assistant to read a company's files or call a business tool usually had to write a custom connector for that one source. Repeat that work for every assistant and every source, and the effort multiplied quickly. Anthropic cast the protocol as an answer to information silos and legacy systems. Earlier fixes had helped only partway. In 2023 OpenAI offered a function-calling API and the ChatGPT plug-in framework. Both tackled similar needs, but each still required connectors tied to a single vendor. What was missing was an open standard that no one vendor controlled.

  5127

• 3. Two Engineers, One Announcement

  ›

  David Soria Parra and Justin Spahr-Summers are the Anthropic engineers who created MCP. The company announced it in November 2024 as an open standard for linking AI assistants to data systems. Its examples included document repositories, business management tools and development environments. The designers did not start from a blank page. MCP reuses the message-flow ideas of the Language Server Protocol, known as LSP, and its messages travel over JSON-RPC 2.0.

  1451110

• 4. What It Does

  ›

  At release, MCP came with software development kits in Python, TypeScript, C# and Java. Anthropic also maintains an open-source repository of reference server implementations and SDKs. Through the protocol, a program gets a standard interface for reading files, executing functions and handling contextual prompts. The specification covers data ingestion and transformation, contextual metadata tagging, and interoperability across different platforms. Connections run in both directions between data sources and AI tools. One everyday use is natural language data access, where a person queries a structured database in plain language. An official extension called MCP Apps, built on mcp-ui, goes further. The base specification is limited to text and structured data. MCP Apps standardizes interactive interfaces such as dashboards, forms and data visualizations, delivered from MCP servers to host applications like Claude and ChatGPT.

  510111

• 5. Developer Territory

  ›

  Replit is one of the coding platforms that adopted MCP, alongside integrated development environments and code intelligence tools like Sourcegraph. Their aim is to give AI coding assistants real-time access to project context. Beyond the coding world, MCP can be integrated with Microsoft Semantic Kernel and Azure OpenAI. MCP servers can also be deployed to Cloudflare. Those integrations set the stage for a bigger step by the very companies that had built their own competing connection methods.

  121575

• 6. Rivals Become Stewards

  ›

  OpenAI and Google DeepMind are among the major AI providers that adopted the protocol after its announcement. OpenAI made its adoption official in March 2025, having already built the standard into its products, including the ChatGPT desktop app. In September 2025, OpenAI added MCP support to ChatGPT apps, which allows third-party access inside ChatGPT. Then, in December 2025, Anthropic donated MCP to the Agentic AI Foundation, or AAIF. The AAIF is a directed fund under the Linux Foundation. It was co-founded by Anthropic, Block and OpenAI, with support from other companies. In April 2026, the AAIF held the MCP Dev Summit North America in New York City, drawing approximately 1,200 attendees.

  2381417

• 7. Praise and Warnings

  ›

  The Verge reported that MCP meets a growing demand for AI agents that are aware of context and able to pull from many different sources. Security researchers were less comfortable. In April 2025 they released an analysis concluding that MCP had multiple outstanding security issues. One was prompt injection. Another was tool permissions that let someone combine tools to exfiltrate data. A third was lookalike tools that can silently replace trusted ones. Some observers have likened MCP to OpenAPI, a similar specification that aims to describe APIs.

  421222926