MCP

MCP, for Model Context Protocol, is an open protocol published by Anthropic in November 2024 then transferred to an open-source foundation in December 2025. Its goal: standardize how language-model-based AI agents (Claude, GPT, Gemini, Mistral) access external tools, data sources and business functions. It is the equivalent of what HTTP was for the web: an interoperability protocol that survives changes of implementation.

Before MCP, every LLM vendor exposed its own tool API (OpenAI function calling, Anthropic tool use, etc.), incompatible with each other. For an agent to switch models, you had to rewrite every connector. MCP unifies: an MCP server exposes its capabilities, and any client (Claude, GPT, custom agent) can connect to it without adaptation.

Why MCP matters in 2026

• Major adoption: the main LLM providers adopted or announced MCP between late 2024 and mid-2025.
• Model decoupling: an agent that speaks MCP can switch from Claude to GPT to Mistral without modifying its connectors.
• Security by construction: exposed capabilities are declared, scoped, authenticatable. No more ad hoc plumbing.
• Discoverability: an agent dynamically discovers the tools available in an MCP server, with no manual documentation.

MCP is relevant to three audiences. SaaS vendors who want their APIs to be consumable by customer AI agents. Internal IT teams building agents who want to stay free in their model choice. CTOs who want to avoid lock-in by an LLM vendor. In 2026, exposing your APIs in MCP becomes a standard equivalent to exposing a REST API in 2015.

At Swoft, MCP is one of five protocols generated isomorphically from the operations registry. A business operation is declared once in the metamodel, and it is automatically exposed as REST HATEOAS, MCP, real-time Server-Sent Events, TypeScript types for the front-end, and a CLI command. The five protocols share the same input schema, preconditions and guidance metadata. No desynchronization is possible.

Concretely, Swoft exposes 33 MCP servers organized along Conway's Law, totaling more than 700 tools. Each AI agent, built by Swoft or by a customer, accesses these tools through the same MCP channel as Claude or GPT. This uniformity reduces integration cost and ensures agents survive model migrations.