Model Context Protocol (MCP)

MCP standardizes how an AI application discovers and calls capabilities that live elsewhere. An MCP server exposes tools, resources, and prompts; an AI application connects to it through a client and makes those capabilities available to the model. Anthropic introduced it in late 2024, and in December 2025 it was donated to the Agentic AI Foundation under the Linux Foundation.

For a product team, MCP matters in two directions. Inbound, your feature can connect to third-party MCP servers so your assistant can work with a customer's issue tracker, drive, or CRM without you writing a custom integration for each. Outbound, you can expose your own product as an MCP server so customers can use it from the AI assistants and coding tools they already use.

Both directions carry real security questions. Connecting a server gives the model new actions and new sources of untrusted text, which is a prompt-injection path. Use per-user credentials, an allowlist of servers and tools, and confirmation for consequential actions. Exposing your product means authentication, permission scoping, and rate limits, just like a public API.

For simple features over your own data, ordinary tool use inside your app is often enough; reach for MCP when integrations need to cross product boundaries.