MCP (Model Context Protocol) is an open protocol that standardizes how applications provide context to large language models (LLMs). It offers a consistent way to connect AI models to diverse data sources and tools, enabling agents and complex workflows that link models to the outside world.
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open protocol that standardizes how applications hand context, data, and tools to large language models. Introduced by Anthropic and quickly adopted across the AI ecosystem, it gives an agent a single, predictable way to reach the outside world instead of a different bespoke integration for every source. I think of it as the connective tissue that turns a model into something that can actually do work against my APIs and data.
- Client-server architecture - An MCP host (an AI application) talks to MCP servers that expose capabilities over a defined transport.
- Tools, resources, and prompts - Servers advertise callable tools, readable resources, and reusable prompts in a consistent, discoverable shape.
- Transport-agnostic - The protocol runs over stdio, HTTP, and streaming transports, so a server can live locally or across the network.
- Built on JSON-RPC - Messages use JSON-RPC 2.0, giving MCP a familiar, well-understood request/response foundation.
In practice, MCP is how the agentic turn actually reaches an API. Across API Evangelist I wrap catalog search, governance checks, and data access as MCP servers so an agent can call them the same way it calls anything else, and I lean on OpenAPI as the source of truth that those servers are generated from. Publishing an MCP server is fast becoming a baseline expectation for API providers who want their capabilities to be usable by agents, not just human developers.
Industry: AI
Also in the data model catalog
This standard shows up on both sides of the house — as a standard applied in API operations here, and as a data model defining what information exists. Same standard, two lenses.
See it as a data model →Open source tools that implement this standard
Harvested from this standard's own governing organisation, with the license read from each repository rather than assumed. The role is what the tool does to the specification, from the shared vocabulary — so an agent holding one of these documents can resolve straight from a goal to a tool.
| Tool | Role | License | What it does |
|---|---|---|---|
| fx | runs |
Apache-2.0 | A tiny, open, native coding agent from Vercel Labs — a roughly 6MB Zig binary that reads code, edits files and runs commands in... |
Referenced on the API Evangelist blog
Where this standard shows up across sixteen years of my writing at apievangelist.com — how it fits into API design, governance, and the agentic turn.
How I Actually Use MCP Across API Evangelist
2026-07-22
Put Governance Where the Agent Can Reach It
2026-07-07
Me Controlling the Protocol (MCP)
2025-03-20
