TECHNOLOGIES

MCP

Model Context Protocol

An open protocol that lets an AI model securely connect to external tools and data sources, without bespoke integrations per model.

Last reviewed: 2026-05-24 byKevin Riedl wiki β†—

Model Context Protocol (MCP) is to AI agents what HTTP is to the web: a shared, open standard for how the model talks to the rest of the world. It was introduced by Anthropic in late 2024 and adopted broadly in 2025 and 2026.

The technical shape: an MCP server exposes resources, tools, and prompts via a defined schema. Any MCP-capable client (Claude Desktop, Claude Code, the API, an IDE plugin) can discover and call those tools without per-vendor glue. Build the integration once; every MCP client benefits.

For businesses the practical implication is leverage. Internal tools, databases, and APIs wrapped in an MCP server become usable by whichever LLM your team is on this quarter, with no rewrite when you switch. The trade-off is that MCP is still young: tooling, security models, and audit patterns are maturing. We build MCP servers regularly and have shipped them to production; we also still write a security review for each one because the standard is moving.

// FAQ

FAQs

FAQs

Function calling is per-vendor. MCP is portable. If you only ever use one model provider, function calling is fine. If you want to be able to swap providers (most enterprises do), MCP is the lower-lock-in choice.
MCP itself is a transport protocol; security depends on what the server exposes and how it authenticates. A poorly-built MCP server is a confused-deputy waiting to happen. A well-built one is no riskier than any other internal API.
Two criteria: the underlying tool is genuinely useful inside an LLM loop, AND the security model permits a model to invoke it (preferably with human-in-the-loop on destructive actions). Anything else stays a regular API.