How to Set Up Cross-Platform Sync for Claude Cowork

Claude Cowork is available on both macOS and Windows, but each installation is independent by default. Files you upload on your work desktop don't automatically appear on your laptop. Upload context is local to that machine and that session.

In practice, this means you can't start a task on one machine and pick it up on another. The session state and uploaded files stay where they were. Moving file access to a cloud layer fixes this: the agent reads from a central repository instead of local storage, so the same files are available on any device.

For teams, the benefit compounds. When everyone's Claude instance reads from the same workspace, you don't get inconsistent answers from agents working off different copies of the same files.

Helpful references: Fast.io Workspaces, Fast.io Collaboration, and Fast.io AI.

When you upload a document to a local Claude session, that file is processed temporarily. It is not shared or versioned. Close the desktop application, and the context is gone. You re-upload files and re-explain the task the next time.

For teams, this compounds. Say an engineering team uses Claude to refactor a codebase. One developer runs the agent locally on a Mac, another on Windows. Even sharing a project, their AI assistants read from different files and give different answers. Version conflicts and repeated context-setting follow.

Consumer cloud storage tools like Dropbox or Google Drive reduce this for humans, but they were not designed for AI agents. They rely on background sync clients that download files locally before an agent can read them. That introduces delays and consumes local storage. When an agent needs to parse gigabytes of documentation, waiting for a sync client creates interruptions. Agents need direct API access to data, not local read-write operations.

Three things need to be true for cross-platform sync to work well with Claude Cowork.

Storage has to live in the cloud, not on local disk. The agent should read and write via API, so access speed depends on network latency rather than local hardware. If the agent has to wait for a sync client to download files first, you've reintroduced the bottleneck you were trying to avoid.

The storage layer needs to understand file contents, not just store them. When an agent looks for a document, it usually needs to search by meaning rather than by exact filename. That means vector embeddings need to be generated for uploaded documents.

Concurrent access handling matters too — especially for teams. If two Claude sessions try to edit the same file at once, you need file locking to prevent conflicts. Connecting via MCP avoids the need for custom middleware to coordinate this.

Fast.io connects Claude to a cloud workspace using its Model Context Protocol (MCP) server. Instead of relying on local sync clients, Claude reads and writes directly from a cloud environment accessible on any device.

When you connect Claude Cowork to Fast.io, the agent gains access to a workspace where every file is automatically indexed. Fast.io's Intelligence Mode handles document parsing, semantic indexing, and RAG natively. The agent queries the workspace and gets cited answers without downloading files first. This reduces the context window load on the LLM and speeds up responses.

Other MCP-compatible storage options exist — MinIO for fully self-hosted setups, or Google Drive via its own MCP server if you want to stay within Google's ecosystem. Fast.io is worth considering when you need persistent agent storage, built-in RAG, and team collaboration in one place, with a free tier available for initial setup.

Fast.io supports shared workspaces for agents and human users. An agent generates a report and saves it to a Fast.io folder; a team member reviews that document on a different operating system with comments and real-time presence. This removes the isolation of local Claude sessions.

Connect Claude to a Shared, Cross-Platform Workspace

Stop re-uploading files on every device. Get a free cloud workspace your Claude agent can access from anywhere.

Sign Up Free

Setting up cross-platform sync requires configuring a cloud workspace and linking it to your Claude Cowork application. Follow these steps to connect your files.

Step 1: Set Up a Workspace Create an organization in Fast.io and set up a dedicated workspace for your AI agent interactions. This workspace acts as the cloud repository bridging your Windows and macOS environments. Invite team members so they can monitor and collaborate on the agent's work.

Step 2: Upload Your Files Move the local files you want the agent to use into this workspace. You can upload directories via the web interface or API. Once uploaded, activate Intelligence Mode. The system indexes documents automatically and generates semantic embeddings, so you skip setting up a separate vector database.

Step 3: Configure the MCP Server Deploy the Fast.io MCP server to connect Claude Cowork with your workspace. Add the Fast.io MCP server endpoint to your Claude desktop application's configuration file. Generate an API key from your Fast.io dashboard and pass it as an environment variable to authenticate.

Step 4: Work Across Platforms Open Claude Cowork on macOS or Windows and authenticate against the same Fast.io MCP server. Issue commands like "Summarize the project requirements in the Q3 planning folder" and the agent retrieves data from the cloud. Files the agent generates save back to the Fast.io workspace and are accessible to other connected devices and team members.

The key architectural principle is that state lives on the server, not in the client. With MCP, the Claude desktop app is just a client — you can close it, restart it, or switch machines without losing continuity because the workspace state is entirely server-side.

File locking handles concurrent modifications. When an agent starts editing a dataset, it holds a temporary lock that prevents another session from overwriting the same file. The lock releases when the operation finishes.

Webhooks are useful for downstream automation. Rather than polling for updates, connected services receive a push notification when a file is added, changed, or deleted. This means downstream jobs — compiling reports, notifying reviewers, triggering builds — fire immediately when the agent saves a change rather than on a timer.

Tying files to individual desktop sessions works fine for personal use. Once you're sharing work across a team, or running multiple agents on related tasks, the per-session model breaks down quickly.

A cloud workspace with MCP also means swapping the underlying model doesn't disrupt your file setup. Your documents stay indexed regardless of whether you're using Claude, switching to a different model for specific tasks, or running multiple agents in parallel. The file layer doesn't care which client is reading it.