Model Context Protocol (MCP)
The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in November 2024, that defines how AI assistants connect to external tools and data sources through a single uniform interface instead of a separate custom integration for every system.
Before MCP, every AI-to-tool integration was bespoke. An assistant that wanted to read a CRM, query a database or change an ad campaign needed a one-off plugin written against that vendor's API, and that work had to be repeated for every client the assistant ran in. MCP replaces that N-by-M problem with one protocol: an AI client talks to an MCP server over a standard message format, and the server advertises what it can do.
The protocol has three primitives. Tools are functions the model can call, such as pausing a campaign or fetching a report. Resources are read-only data the model can pull into context. Prompts are reusable templates. Servers run either locally as a process on your machine or remotely over HTTPS. In both cases your credentials sit with the server, not the model — the server decides what is allowed and what is refused.
Because MCP is open, a server written once works in Claude, in ChatGPT, and in any other MCP-compatible app. But the protocol describes the plumbing, not the safety policy. An MCP server can be strictly read-only or it can make permanent changes to a live account; that is a choice made by whoever built the server. So the useful questions about any MCP server are: which tools does it expose, does it write, and does it ask before it writes.
Opus Growth is a hosted remote MCP server for advertising. You paste one URL — mcp.opus-growth.com — into your AI app, sign in with Google, and you are connected in about 60 seconds: no developer token, no API keys, no self-hosting, no ETL. That opens a 221-tool catalogue, and these surfaces are live today: Google Ads (69), Microsoft Advertising (30), TikTok Ads (25), LinkedIn Ads (18), Google Tag Manager (18), GA4 (10), Search Console (6) and YouTube (5). Meta Ads (27 tools) is in final platform review and Business Profile (7) is waiting on Google's API quota approval, so neither is live yet; both open the moment approval lands.
Frequently asked questions
Anthropic introduced and open-sourced MCP in November 2024. The specification and SDKs are public, and the protocol is now supported by multiple AI apps including Claude and ChatGPT, so the same server can serve all of them.
No. An API is the underlying service that does the work; MCP is a standard way of describing that service's capabilities so a model can discover and call them. An MCP server typically sits in front of one or more existing APIs.
Both are possible and it depends entirely on the server. Read-only servers just report; write-capable servers can create, update or delete records. For write-capable servers, check whether they show a dry-run preview and require your approval first.
