What is MCP, and why does it matter?

Model Context Protocol is an open standard for connecting an AI client to tools and data, so one integration works across Claude, Cursor, Codex and anything else that speaks it.

It replaces one bespoke integration per AI client with one server many clients can use.
It is a transport and a tool-description format, not a permission system.
Whoever runs the server still has to decide what the tools are allowed to do.

The problem it solves

Before a shared protocol, giving five AI clients access to the same internal data meant five integrations, five auth flows and five things to keep current.

MCP defines how a client discovers what tools a server offers and how it calls them, so the server is written once and every compatible client can use it.

What it does not solve

MCP does not decide what a tool should be allowed to do. It describes the tools and carries the calls; the judgement about which of them a given person may invoke, and whether a write should stop for approval, sits with whoever runs the server.

This matters because the interesting failure is not a malformed call, it is a well-formed call that should not have been made. In Indexbrain the MCP surface is deliberately read-first: of 32 tools, only four write, and no MCP tool can run a skill at all.

How it is used in practice

Claude, Cursor, Claude Code and Codex connect to a remote MCP server with OAuth. ChatGPT does not use MCP for this and connects through a REST API and custom actions instead.

You can also point a client at your own MCP server. Indexbrain supports adding one over HTTP, SSE, or a sandboxed command, and its tools are permission-gated exactly like every built-in one.

Related questions

Give your AI a company to work for

Connect your tools, and every agent you already use starts the day knowing how your company actually works.