Model Context Protocol (MCP)

What it is

Before MCP, every link between an assistant and an external system was bespoke, and the same work was repeated for each client. MCP reduces this to one protocol: the client connects to the server with a standard message format, and the server announces what it can do.

There are three building blocks. Tools are functions the model can call; Resources are read-only data it can bring into context; Prompts are ready-made templates. The server runs as a local process on your machine or remotely over HTTPS; either way your credentials live on the server, not in the model.

Why it matters

Because it is an open standard, a server written once works in Claude, ChatGPT, Claude Code, Cursor and any other MCP-capable app. You do not have to connect your ad account to each assistant separately.

The protocol only describes the pipe, not the security policy. An MCP server can be entirely read-only or make permanent changes in your live account. Three questions to ask: which tools exist, does it write, and does it ask for approval before writing?

How Tecrube uses it

Tecrube is a remote MCP server built for ad operations. You paste the address into your app and sign in with Google; about 60 seconds. No developer token, no API key, no server of your own.

Behind it sit tools for Google Ads, Meta Ads, Microsoft Advertising, TikTok Ads and LinkedIn Ads, plus Google Tag Manager, GA4, Search Console, YouTube and Business Profile. Every write action passes through the preview and approval gate and lands in the audit log.

Example

You type “show me the last 7 days” in Claude; the client calls Tecrube's report tool, the server fetches the data from the Google Ads API, and the model summarises the numbers. No API key ever passes through the chat.