An open standard that defines how AI systems can connect to external tools, data sources, and services in a consistent, secure way. Introduced by Anthropic in late 2024 to solve the N×M integration challenge.
Model Context Protocol (MCP) is an open standard introduced by Anthropic in late 2024 that solves a fundamental problem: every AI assistant previously needed custom connectors or plugins for each separate data source or API, leading to a combinatorial "N×M" integration challenge.
MCP replaces these fragmented, one-off integrations with one standardized protocol, so developers can build a connector once and any MCP-compatible AI app can use it.
Modern AI assistants are powerful at language tasks, but without external context or tools, they have significant blind spots. They cannot:
Before MCP, connecting AI to these capabilities required vendor-specific solutions or custom code for each tool needed. This led to duplicated effort and brittle systems where every developer was writing their own adapters for Google Drive, Slack, GitHub, etc.
MCP enables AI assistants to access fresh, relevant information and perform actions in the real world, rather than being limited to their training data.
Built on JSON-RPC 2.0 with proven transport protocols (HTTP, STDIO), following the same patterns as Language Server Protocol (LSP).
Requires explicit user approval before any tool is executed, ensuring humans remain in control of impactful operations.
MCP follows a client-server architecture on top of an AI assistant (the "host"):
Tools are operations the AI can request - like "send an email," "query a database," or "schedule a calendar event." Each tool has a JSON Schema defining expected inputs and outputs.
Resources provide read-only data that AI can fetch as context - files, database entries, or any structured data relevant to the AI's task. Resources use URI identifiers and MIME types.
Prompts are predefined, parameterized interaction templates that streamline complex workflows - like multi-step scripts that can be invoked by name.
Major AI providers have announced support for MCP:
Configure MCP servers in VS Code settings to let Copilot access external tools:
Run local MCP servers for enhanced capabilities:
Companies like Block (Square) use Claude with custom MCP connectors to internal databases, enabling employees to query proprietary data securely through AI assistants.
MCP builds on established standards:
Ready to explore MCP? The protocol is designed to be accessible:
MCP represents a new era of AI connectivity - transforming AI assistants from static Q&A bots into agents that can carry out tasks with real-world context and effects, all while maintaining user control and safety.