What does MCP stand for?

MCP stands for Model Context Protocol — an open standard for connecting AI models and agents to external tools, data, and context sources.

What is an MCP server?

An MCP server is a program that exposes tools, resources, or prompts over the Model Context Protocol. An AI client (like Claude Code or Cursor) connects to it and can call those tools during a task.

How is MCP different from RAG?

RAG retrieves documents and stuffs them into a prompt. MCP is a live protocol: the agent calls tools on demand and can also take actions, not just read text. They're complementary.

MCP (Model Context Protocol)

MCP is an open protocol that lets AI applications connect to external tools and data through one standard interface — agents fetch context and act on demand.

The Model Context Protocol (MCP) is an open standard that gives AI applications a uniform way to connect to external systems. Instead of building a bespoke integration for every model-to-tool pairing, a developer exposes an MCP server, and any MCP-speaking client — such as Claude Code or Cursor — can discover and call its tools.

Why it matters for coding agents

A coding agent is only as good as the context it can reach. MCP lets an agent pull context on demand rather than depending on whatever was pasted into the prompt. For example, a get_product_context tool can return the approved decisions relevant to the current task — so the agent starts already knowing the constraints instead of rediscovering them one broken pull request at a time.

Core concepts

Tools — functions the agent can call (e.g., fetch decisions, run a query).
Resources — data the server exposes for the model to read.
Prompts — reusable prompt templates the server offers.

How an MCP call works, in one example

MCP messages are JSON-RPC 2.0. The pieces are a host (the AI application, such as an editor or CLI), a client inside that host that holds the connection, and a server that offers capabilities. A typical exchange:

The client connects and asks the server which tools it offers.
The server lists them, each with a name, a description, and an input schema, for example get_product_context taking a short task description.
Mid-task, the model decides it needs that tool. The host asks the user for permission if required, then the client sends the call.
The server returns a result, and the host adds it to the model's context window for the next step.

The spec also lets clients offer features back to servers, such as elicitation, where a server asks the user for missing information. The current version is published at modelcontextprotocol.io.

MCP vs an API vs RAG

An API is one service's interface, and every model integration with it is custom. MCP is a shared wrapper format, so one server works with any compliant client. RAG is a retrieval technique, not a protocol; a RAG pipeline is often exposed to agents as one MCP tool. The longer comparison is in MCP vs RAG.

How Lockstep uses MCP

Lockstep exposes an MCP server so agents can fetch the team's approved decisions before they write code, through the get_product_context tool. Separately, a pull request check flags code that conflicts with an approved decision (it warns by default; blocking is opt-in). For persistent context across sessions, see persistent memory for AI coding agents. Read more in context engineering and The Feed Is Coming for GitHub.