Dropbox Dash MCP Server

The Dropbox Dash MCP Server opens up Dropbox Dash search to AI assistants. Dropbox Dash is an AI-powered search for teams that pulls content from many work tools into one place. According to Dropbox, you can connect Dropbox, Google Drive, Microsoft OneDrive, Gmail, Outlook, Slack, Microsoft Teams, Notion, Confluence, Jira, Asana, GitHub, Salesforce, Airtable, Miro, Canva and Zoom, among others. MCP, the Model Context Protocol, is an open standard that lets assistants such as Claude, ChatGPT, Codex or Cursor reach external services. The server comes from Dropbox itself and runs on its infrastructure at mcp.dropbox.com/dash. According to the help page, it is available to all Dash users.

What is the Dropbox Dash MCP Server good for?

Company knowledge rarely sits in one place. The decision lives in a Slack thread, the spec in Confluence, the quote in Google Drive and the bug report in Jira. If you wanted an AI assistant to answer a question about it, you had to dig up the right documents yourself and paste them into the chat. With the Dash server, the assistant does the searching. You can ask what was discussed with a particular client last quarter, where the latest version of a draft contract lives, or which tickets are still open for a feature. The assistant searches across every connected source, reads the hits and answers with reference to the documents it found. Developers benefit as well: Claude Code or Codex can pull in background from tickets, wikis and chats without you setting up a separate MCP server for every tool.

The tools at a glance

According to the help page, the hosted server offers eleven tools. With dash_search, the assistant searches all connected sources and can filter by file type, source and source-specific labels. Several read tools then return a document's metadata and text, content as Markdown in sections, page images of PDFs or slides, and raw data for images, audio and video. Other tools list your connected sources, show sample queries for a source type, resolve links into Dash identifiers and report which account is signed in. Two more tools run search actions directly through a connected source's own API, again read-only.

Read-only: a strength and a limit

The Dash server does not change anything. It creates no files, moves nothing and sends no messages. That lowers the risk considerably, especially when the assistant reads content from many sources. If you want to organize, share or restore files in Dropbox, you need the separate Dropbox MCP Server at mcp.dropbox.com/mcp instead. You can connect both side by side. Through Dash, the assistant also only sees what your account is already allowed to see in the connected apps.

Hosted or local

Besides the hosted server, Dropbox publishes a local Dash server on GitHub under the Apache 2.0 license. It runs on your machine with Python 3.10 or later and the uv package tool, talks over STDIO only and offers four tools: two for sign-in, one search and one detail lookup. According to the README, it trims longer content to roughly 20,000 characters. On the help page, Dropbox explicitly recommends the hosted server, because the local one is updated less often.

Who is the Dropbox Dash MCP Server for?

The server pays off for teams that already use Dropbox Dash and want to tap their scattered knowledge from Claude, ChatGPT, Codex or Cursor. Without Dash access it is of no use, since it searches exactly the index Dash builds from your connected apps. On cost: Dropbox lists no separate price for the MCP server itself. According to Dropbox, Dash comes in paid plans such as Dash for Teams and Dash for Business. Team admins can start a trial, which the help page currently limits to the US, the UK and Australia with an English-language account. Current terms are on Dropbox's pricing pages. Depending on the assistant, you may also need a plan there that supports custom connectors or developer mode.

A Dropbox account with access to Dropbox Dash and apps connected there. Plus a supported client: Claude Code or the Codex CLI installed locally, Claude with access to custom connectors or ChatGPT with developer mode (each only on certain plans), or Cursor. In Dropbox teams, the admin must allow connecting apps. The local server additionally needs Python 3.10 or later and uv.

The Dropbox Dash MCP Server lets your AI assistant search Dropbox Dash, meaning Dropbox and every app you have connected in Dash. It runs at Dropbox under https://mcp.dropbox.com/dash and is read-only. This guide shows the setup in the supported clients and what to watch out for afterwards.

Step 1: Check first

You need access to Dropbox Dash. Check which apps are connected in Dash, because the assistant only finds what is indexed there. In Dropbox teams, the admin may restrict connecting apps. For Claude in the browser or desktop app, you need a plan with custom connectors; for ChatGPT, a plan with developer mode.

Step 2: Claude Code

Run claude mcp add --transport http DropboxDashMCP https://mcp.dropbox.com/dash. On first use, your browser opens and you allow access to your Dropbox account. Inside a session, /mcp shows whether the server is connected.

claude mcp add --transport http DropboxDashMCP https://mcp.dropbox.com/dash

Step 3: Codex

Add the server with codex mcp add DropboxDashMCP --url https://mcp.dropbox.com/dash. Then sign in with codex mcp login DropboxDashMCP.

codex mcp add DropboxDashMCP --url https://mcp.dropbox.com/dash
codex mcp login DropboxDashMCP

Cursor

Open Settings, then Cursor Settings and Tools and MCP. Click New MCP Server and add an entry Dropbox Dash MCP under mcpServers with the url field set to https://mcp.dropbox.com/dash. After saving, Cursor asks you to sign in to Dropbox.

Claude and ChatGPT in the browser

In Claude, go to Settings, Connectors and choose Add custom connector. Enter Dropbox Dash MCP as the name and the server URL as the address. In ChatGPT, first enable Developer mode under Settings, Apps, Advanced settings. Then go to Settings, Apps, click Create app, enter the server URL, choose OAuth and complete the sign-in.

Other MCP clients

Only the five clients above register with Dropbox automatically. For everything else, create your own app in the Dropbox App Console with Scoped access and Full Dropbox. Enable the scopes account_info.read, dash/content.read and dash/content.write and add your client's redirect URIs. Set up the client with the app key, app secret and a generated access token. The token is short-lived and has to be regenerated regularly.

Step 4: The local server as an alternative

Dropbox also publishes a local server on GitHub under the Apache 2.0 license. Clone the dropbox/mcp-server-dash repository, run uv sync, set the APP_KEY environment variable and start it with uv run src/mcp_server_dash.py. It needs Python 3.10 or later and offers only four tools. Dropbox recommends the hosted server, because the local one is updated less often.

uv run src/mcp_server_dash.py

Search well, work safely

Feel free to name the source or file type in your questions, such as only Slack or only PDFs. The assistant can filter on that and gets to the right result faster. Keep in mind that content read from emails or chats is passed to the AI provider. If you have also connected servers with write access, approve their actions one by one, since a document found in search can carry hidden instructions too.

Dropbox OAuth. For Claude Code, Claude web and desktop, ChatGPT web, Codex and Cursor, the client registers itself automatically (Dynamic Client Registration). Other clients need their own app in the Dropbox App Console with an app key, app secret and a short-lived access token. The local server signs in via OAuth through two dedicated tools and stores the token in the system keyring.

Read access to all content your account may see in Dropbox Dash and the apps connected there. For custom apps, Dropbox lists the scopes account_info.read, dash/content.read and dash/content.write with Full Dropbox access; according to the help page, the tools themselves are read-only.

The hosted server runs at Dropbox. Content the assistant reads, including from Slack, Gmail or other connected apps, is passed to the respective AI provider and handled under that provider's terms. The local server runs on your machine and talks directly to the Dropbox APIs.

Low to medium. The server cannot change anything, but it reads content from many sources and passes it to the AI provider. Confidential material from emails or chats can end up in the conversation that way. A crafted document with hidden instructions can also sway the assistant when other servers with write access are connected.