MCP stands for Model Context Protocol. The name sounds technical, and most of what's been written about it is aimed at developers. The underlying concept is simple, and the practical use case for marketers is one of the clearest applications of the technology.
Here's what it does: MCP lets AI assistants connect to external tools and data sources, so they can read from those sources during a conversation rather than only knowing what's in their training data or what you've manually pasted in.
Before MCP, if you wanted your AI to reference something specific — your brand guidelines, your past campaigns, your saved examples of copy that works — you had to paste it into the conversation yourself. Every session. Every time. And the context window is limited, so most of what you'd want to include never made it in.
MCP makes the connection persistent and automatic. An AI assistant with an MCP connection to a tool can query that tool directly, search it, and reference it without you having to do anything.
Why this matters for a swipe file specifically
A swipe file has a fundamental usability problem: you have to remember to open it.
You're writing a brief. You want something punchy, a particular register you've seen done well before. You know you've saved examples of this. But to use them, you'd need to open your folder, search for what feels relevant, pick two or three examples, paste them into your prompt as context, and then ask your AI to write. Nobody does that. It interrupts the work and it's slower than just writing the prompt and hoping for the best.
MCP removes that friction. When your swipe file is connected to your AI through MCP, your AI can search the collection itself before writing — finding the relevant examples, pulling them into context, and working from them without you having to locate and paste them manually.
The collection you've spent years building becomes something your AI actively uses, rather than a resource you intend to reference and usually don't.
What the connection actually looks like
The Kaleidoscope MCP setup takes one paste. From your settings, you copy a connection prompt that includes your personal server URL and authorization token. You paste it into Claude Code. Your AI reads the setup, confirms the connection, and lists the tags currently in your library.
From that point on, Claude Code can search your swipe file, retrieve specific cards, and reference your collection in any creative task. You can also add a note to your CLAUDE.md file so the connection persists automatically across sessions — your AI knows to check your library before any marketing or creative work without being asked.
The technical infrastructure (an MCP server at a personal URL, a Bearer token for authentication) runs entirely on Kaleidoscope's side. From your side, it's a paste.
MCP for marketers vs. MCP for developers
Most MCP content is written for developers, because most early MCP use cases are developer-facing: connecting Claude to a code repository, querying a database, reading from a file system. The developer use cases are real and well-documented.
The marketer use case is less documented and, in some ways, more immediately valuable. Developers often have access to their tools programmatically already. Marketers typically don't — their creative reference material lives in folders, bookmarks, and Notion pages that no tool can query automatically.
An MCP connection to a structured swipe file solves a problem that affects every marketer using AI for creative work: the disconnect between the examples that inform their taste and the AI that's doing the writing.
The broader context
MCP was published as an open standard by Anthropic in late 2024. It's now supported by a growing number of AI tools and platforms. The idea — that AI assistants should be able to read from external data sources rather than operating from training data alone — addresses one of the practical limitations of working with AI on real-world tasks.
For creative work specifically, the limitation it addresses is one that advertising professionals identified decades before AI existed: good work requires good raw material, and the raw material needs to be organized and accessible at the moment you need it. MCP is what makes that accessible to an AI rather than just to you.