Model Context Protocol (MCP) is an open standard that gives AI applications a secure, consistent way to connect with external information and tools. An MCP server lets an AI tool find allowed content, use the actions it supports, and return results—without having to add a different custom integration for each service.
Whatever your industry, you’re no doubt being encouraged to experiment with using AI to speed up daily tasks. But how do you shift from a few early experiments to reliable workflows your whole team can repeat? Connect AI to the content, permissions, and tools people use every day, and you can close that gap. That’s where an MCP server comes into play.
Dropbox has an MCP server that allows you to securely connect tools like ChatGPT to your files. Learn more about the Dropbox remote MCP server and connect your favorite AI tools to the files you use every day—simply choose a plan to get started.
What is MCP?
MCP stands for Model Context Protocol. It defines a shared way for an AI application to discover what an external service can provide, request relevant context, and ask that service to carry out supported actions.
An AI model doesn’t automatically know what’s inside your private project folder or have permission to create a shared link. However, a connected MCP server can make selected information and actions available to the AI application in a standard format.
MCP can support two broad types of work:
- Using context—such as finding a document, reading supported file content, or checking file details
- Taking action—such as creating a folder, organizing files, or generating a link when the connected service supports it
MCP provides a standard way for an AI client to access permitted data and do certain actions—it doesn’t replace the permissions and controls of the connected service. That means, if you connect to the Dropbox MCP server, all your existing file permissions and settings still apply.
What is an MCP server?
An MCP server is a software program that provides selected context and capabilities to an AI application. It may run on your computer or as a remote service, so “server” doesn’t necessarily mean a separate physical machine.
The following three parts work together:
- MCP host—the application you use, such as an AI assistant or coding environment
- MCP client—the component inside the host that manages its connection to a specific server
- MCP server—the program that exposes data and supported actions from an external service
A host can manage several clients, while each client communicates with one server. This separation keeps each connection focused and easier to control, which is crucial for the safe use of AI tools.
How does MCP work?
MCP passes structured requests, such as a prompt, between an AI application and a connected server. Consider this request as an example:
“Find the latest project brief in Dropbox, summarize the changes, and create a link I can share.”
Here’s what happens:
- You make the request—the AI application interprets what you want to find and do with your files
- The host checks available capabilities—its MCP client identifies what the connected Dropbox server can provide
- The client sends the request—it asks the server for the permitted file information or to do the action you request
- The server works with Dropbox—it uses the authenticated connection and existing permissions to find the file, retrieve supported content, or create the link
- The assistant returns the result—it gives you the summary and, when supported, confirms the completed action
The host manages the conversation, the client manages the connection, and the server provides a defined route to the external service.
What can an AI assistant do with files stored in the cloud?
A connected AI assistant can use the files in your cloud storage as context and perform supported file actions. What it can do depends on the AI application, MCP server, connected service, and your permissions. With the Dropbox remote MCP server, possible tasks include:
- Finding files and folders by name or content
- Reading or summarizing supported file content
- Checking file details and sharing information
- Creating folders or organizing existing content
- Creating shared links and file requests
- Saving supported text-based output to Dropbox
- Using project specifications and documentation as context for coding work
This makes it faster to go from looking for a file to getting work done. With an MCP server, you can ask an assistant to locate source material, help you understand it, and prepare for the next step—without switching between tools. Exact capabilities vary by integration.
Is MCP secure?
MCP standardizes the connection, but the protocol alone doesn’t make every server or action safe. Security depends on the host, server, connected service, and the permissions you approve. Consider the following before connecting to an MCP server for extra security:
- Check who operates it and use an official server when one is available
- Review the access requested during authentication
- Grant only the permissions needed for the work
- Review sensitive changes before they run
- Follow any controls set by your team admin
The Dropbox remote MCP server uses OAuth for authentication. In the Dropbox integration for Claude, existing file permissions and admin controls still apply. Claude can only access content and complete actions that the connected person already has permission to use in Dropbox.
Use MCP with Dropbox across the AI tools you choose
With Dropbox, you can safely bring permitted files and folders into the AI environment that fits the task, then keep useful output connected to the work that follows. Here are a few ways to use it:
- Dropbox remote MCP server: Use ChatGPT Web to find, preview, organize, and share Dropbox files. You can also connect to Codex to use project files as technical context, or Cursor to view and organize content, move or copy files, make links or file requests, and create text-based files.
- Dropbox connector for Claude: Bring supported Dropbox content into Claude conversations to preview, find, search, and share files—then save to Claude-generated content direct to Dropbox. Turn AI output into work your team can keep using.
- Dropbox plugins for Claude Cowork and Claude Code: Work with permitted Dropbox content inside collaborative and technical Claude workflows. Keep trusted project context close and move from insight to action without breaking your flow.
Capabilities aren’t identical across every experience, so start with the integration that matches your task. Learn more about the AI integrations available in Dropbox.
MCP servers for secure AI tools
MCP gives AI applications a standard route to external context and actions. With permission-aware Dropbox connections, your AI tools can work securely from the files you already rely on.
The output can return to Dropbox, where you can organize it, share it, review it, and keep work moving after the conversation ends. Choose a Dropbox plan to get started.
Frequently asked questions
No. An API defines how software interacts with a particular service. MCP gives AI applications a common structure for discovering and using capabilities from different services. However, an MCP server may call APIs behind the scenes.
No. JSON is a lightweight text format used to structure and exchange data, which is why MCP messages can look like ordinary JSON. MCP is built on JSON-RPC, but it also defines the roles of hosts, clients, and servers, along with capability discovery, requests, responses, and security boundaries.
Retrieval-augmented generation, or RAG, usually focuses on finding relevant information and adding it to a model’s input. MCP provides a standard connection that can expose information and supported actions. An application can use both MCP and RAG.
ChatGPT can connect to remote MCP servers. The Dropbox remote MCP server currently supports ChatGPT Web and Codex, along with Cursor, Claude Web, and Claude Code. Availability and setup requirements can vary by client and plan.
