dropbox-mcp
Dropbox MCP server to recover deleted files, list revisions, search content, and force-download cloud-only files via server-side API.
README
dropbox-mcp
An MCP (Model Context Protocol) server that exposes the Dropbox API as tools for LLM agents. Built on FastMCP and dropbox-sdk-python.
Focus: recovery and discovery on an existing Dropbox account — restoring deleted files, listing revisions, searching content, and force-downloading cloud-only files. The server talks to the Dropbox server-side API, not the local sync folder, so it can see and restore files that local sync has already deleted.
Tools
All tool names are prefixed dropbox_ to avoid collisions with other MCP servers.
| Tool | Behavior | Read-only |
|---|---|---|
dropbox_restore |
Restore the most recent server-side revision of a deleted file. | No |
dropbox_restore_batch |
Restore multiple files in one call; reports per-path result. | No |
dropbox_restore_revision |
Restore a specific revision ID (e.g., a known-good earlier version). | No |
dropbox_download |
Force-download a file from Dropbox to the local sync folder, bypassing Smart Sync cloud-only state. | No |
dropbox_search |
Search by filename or content across the account. Returns path, size, modified date. | Yes |
dropbox_list_deleted |
List deleted entries in a folder (optionally recursive). Input for restore workflows. | Yes |
dropbox_file_info |
Return size, modified time, revision ID, and content hash for a path. | Yes |
dropbox_list_revisions |
List up to 100 revisions of a file with rev ID, size, and modified time. | Yes |
Read-only tools are safe to call freely. The four restore/download tools mutate state on Dropbox servers or local disk.
Installation
Prerequisites
- Python 3.10 or newer
- A Dropbox account and a Dropbox app
with
files.content.read,files.content.write, andfiles.metadata.readscopes
Install
git clone https://github.com/danielsimonjr/dropbox-mcp.git
cd dropbox-mcp
pip install -e .
Or, for dependencies only:
pip install -r requirements.txt
Authentication
The server loads credentials from ~/.claude/channels/dropbox/.env on startup. Create
that file and paste in the template below, then fill in your values:
mkdir -p ~/.claude/channels/dropbox
touch ~/.claude/channels/dropbox/.env
Template:
# --- Option A: OAuth 2 refresh token (recommended) ---
# Create an app at https://www.dropbox.com/developers/apps, enable the scopes
# files.content.read, files.content.write, files.metadata.read, then run the
# OAuth flow once to obtain a refresh token.
DROPBOX_REFRESH_TOKEN=
DROPBOX_APP_KEY=
DROPBOX_APP_SECRET=
# --- Option B: Long-lived access token (fallback) ---
# Leave blank if you are using Option A above.
DROPBOX_ACCESS_TOKEN=
# --- Optional ---
# Local Dropbox sync folder. Used by dropbox_download to write files.
# Defaults to ~/Dropbox if unset.
# DROPBOX_LOCAL_PATH=C:\Users\you\Dropbox
Two auth modes are supported, tried in order:
- OAuth 2 refresh token (recommended): set
DROPBOX_REFRESH_TOKEN,DROPBOX_APP_KEY, andDROPBOX_APP_SECRET. Access tokens are refreshed automatically, so credentials do not expire. - Legacy long-lived access token (fallback): set
DROPBOX_ACCESS_TOKENonly. Simpler to get from the app console, but tokens expire after a few hours for newer apps.
Running the server
Directly (for testing)
python server.py
The server communicates over stdio, so there is no interactive output — it waits for MCP protocol messages on stdin.
With the MCP Inspector
npx @modelcontextprotocol/inspector python server.py
This opens a browser UI where you can list tools and call them manually.
Registering with Claude Code
Add an entry to your ~/.claude/.mcp.json:
{
"mcpServers": {
"dropbox": {
"command": "python",
"args": [
"-X", "utf8",
"C:\\path\\to\\dropbox-mcp\\server.py"
]
}
}
}
The -X utf8 flag is recommended on Windows so that non-ASCII paths in results do
not trip up the default cp1252 encoding.
Restart Claude Code for the registration to take effect.
Examples
Restore a file deleted by accident:
Agent: dropbox_restore(path="/Projects/report-final.docx")
Result: Restored: /Projects/report-final.docx (rev: abc123, size: 45678 bytes)
Find a file without knowing its exact location:
Agent: dropbox_search(query="RSP consciousness paper", max_results=5)
Result: Found 3 results for 'RSP consciousness paper':
1.25 MB 2026-02-18 /Misc/Philosophy/Beyond the Bat/paper.pdf
0.31 MB 2026-02-10 /Misc/Philosophy/Beyond the Bat/drafts/outline.md
...
Roll back to a specific earlier revision:
Agent: dropbox_list_revisions(path="/report.docx", limit=5)
Agent: dropbox_restore_revision(path="/report.docx", rev="0123abc")
Security notes
- The
.envfile holds long-lived credentials — keep it out of version control (the default.gitignorealready excludes.envfiles). - Restore and download tools are not idempotent from the user's perspective:
they mutate Dropbox state or overwrite local files. Agents should confirm
intent before invoking them, especially
dropbox_restore_batch. - The server binds to no network ports — communication is stdio only, so there is
no inbound attack surface from the MCP layer itself. The only outbound
connection is to
api.dropbox.comvia HTTPS. - Logs go to stderr, never stdout (stdout is reserved for MCP protocol frames).
Development
# Syntax check
python -m py_compile server.py
# Quick smoke test (lists registered tools)
npx @modelcontextprotocol/inspector python server.py
For changes to the tool surface, update both this README and CHANGELOG.md in the
same commit.
License
MIT — see LICENSE.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.