MCP (Model Context Protocol)
An open protocol, created by Anthropic and now hosted by the Linux Foundation's Agentic AI Foundation, that standardizes how AI applications connect to external tools, data sources and services.
Model Context Protocol (MCP) is an open protocol that Anthropic released in November 2024 to standardize how AI applications connect to external data and tools. Before MCP, every integration needed custom glue code for each AI app. MCP defines one interface, so any MCP client (an AI app such as Claude, ChatGPT, Cursor or VS Code) can connect to any MCP server without a bespoke adapter. In December 2025 Anthropic donated MCP to the Agentic AI Foundation, a Linux Foundation fund it co-founded with Block and OpenAI, so the protocol is now developed under vendor-neutral governance.
The architecture
MCP uses JSON-RPC 2.0 messages between a host (the AI application), the clients it runs, and servers. Servers offer three core features:
- Resources: Context and data for the user or the model, such as a file in your repo, a database row or a calendar event.
- Tools: Functions the model can invoke, such as running a SQL query, creating a GitHub issue or sending a Slack message.
- Prompts: Templated messages and workflows that the server provides for users.
Clients can offer elicitation, which lets a server ask the user for more information mid-task. Sampling and roots, client features from earlier versions, were deprecated in the 2026-07-28 revision. Optional extensions add more: Tasks for long-running asynchronous operations, and MCP Apps for interactive UI elements such as charts and forms rendered inside the conversation.
Communication runs over local stdio or, for remote servers, Streamable HTTP, which replaced the original HTTP plus Server-Sent Events transport in the March 2025 revision. The current specification, version 2026-07-28, made the protocol stateless: it removed the initialization handshake and protocol-level sessions, so each request carries its own protocol version and client capabilities. The spec says hosts must get explicit user consent before invoking any tool; each client decides how to ask, from per-call approval prompts to saved permissions.
Why it matters for developers
MCP gives an AI app access to your filesystem, databases, APIs, browser and internal tools through one standard interface. Instead of pasting context into a chat window, an agent can read your codebase, check your CI status and open a pull request directly. Support now reaches well beyond Anthropic: ChatGPT (through custom MCP plugins), OpenAI's Responses API, Cursor and VS Code all act as MCP clients, and services such as GitHub, Notion and IFTTT publish their own MCP servers. In Claude, remote MCP servers show up as connectors.
Building an MCP server
The MCP project maintains official SDKs for TypeScript, Python, C#, Go, Rust, Ruby, Java, Swift, PHP and Kotlin. With version 2 of the Python SDK, a minimal server looks like this:
from mcp.server import MCPServer
mcp = MCPServer("my-server")
@mcp.tool()
def query_db(sql: str) -> str:
"""Run a read-only SQL query and return the rows."""
return run_query(sql)
if __name__ == "__main__":
mcp.run(transport="stdio")The project keeps a small set of reference servers (Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time). Its older GitHub, PostgreSQL, Puppeteer and Slack servers are archived, and vendors now maintain their own, such as GitHub's official GitHub MCP Server. The MCP Registry at registry.modelcontextprotocol.io lists published servers. Because MCP servers can take real actions, treat third-party servers like any dependency: review the tools they expose, treat tool descriptions from servers you don't trust as untrusted input, and keep approval prompts on for anything that writes or sends.
Related terms
Models relevant to MCP (Model Context Protocol)
Claude Opus 5.5
Anthropic's current Opus and its recommended starting model, priced at $4/$20 per 1M tokens
View model →Claude Haiku 5.5
Anthropic's fastest, lowest-cost current model at $0.10 per million input tokens with a 1M context window
View model →GPT-6.1 Sol
Near-Astra intelligence for a fifth of the price
View model →