Give a compatible MCP client tools for working with your cloud files. Files for AI provides project-scoped access for discovery, reading, downloads and versioned writes through a remote endpoint.
Original files. Project permissions. Access you control.
Your files
Originals preserved
Your AI assistantAccess by permission
You decide what’s shared
Only the projects you chooseWORKFLOW ILLUSTRATION
Keep original filesChoose project accessConnect your assistant
BUILT AROUND YOUR FILES
A useful home for the files your AI needs.
Connect with scoped authorization
Sign in and approve specific projects and read or write access. Requests are checked against current membership and connection permissions. A Viewer can authorize their own read-only connection without becoming a project administrator.
Retrieve useful file context
Clients can list files, search names and supported extracted text, inspect metadata, and read bounded text excerpts. Responses identify the version used and report when content is unsupported or a result is truncated.
Choose the right upload path
An agent with HTTP access can upload binary files through signed multipart URLs. A web assistant can submit generated UTF-8 text through the bounded write_file tool. Both paths use the same project permissions and file verification.
Handle changes explicitly
File metadata includes stable file IDs, immutable version IDs and raw SHA-256 hashes. An update must identify the version it expects to replace, so another user's newer revision produces a conflict rather than a silent overwrite.
HOW IT WORKS
From your files to your next conversation.
01
Add the remote endpoint
Configure your compatible MCP client with the Files for AI server address for your workspace. A hosted connection uses HTTPS and the remote Streamable HTTP transport.
02
Choose projects and permissions
Complete sign-in and review the requesting client. Grant read-only access for retrieval, or read/write access when the assistant should be allowed to save files.
03
Discover, read and save
Ask the assistant to find a file and inspect the relevant version. For an upload or generated file, check its verification status before treating the new revision as ready.
CAPABILITIES & LIMITS
Clear about what your files can do.
Storing an original file and reading its contents are different capabilities. Here’s what to expect.
Client support varies. Connector-directory availability and support for a specific assistant should be checked separately.
The remote connector accesses authorized cloud files. It does not grant access to local folders on your computer.
A FEW PRACTICAL ANSWERS
Questions, answered.
What does MCP add to file storage?
MCP gives compatible assistants a common way to discover and call file tools. Files for AI supplies the storage, project permissions and version checks behind those tools. The client still needs to support the remote transport and authorization flow.
Can a web assistant upload a chat attachment with a signed URL?
Only if that client exposes the attachment's bytes and can perform the required HTTP upload. A signed URL does not transfer an attachment by itself. Use the web upload flow when those capabilities are unavailable, or write_file for generated text within its size limit.
What can an agent learn from a file hash?
It can compare the cloud version's verified SHA-256 with a hash it computes from another file. Matching hashes establish byte equality. They do not establish permission, create a persistent link, or authorize automatic updates.
What happens when I revoke a connection?
Subsequent connector requests lose access. A storage URL issued before revocation can remain usable until its short expiry, so revoking a connection does not promise immediate invalidation of every previously issued download URL.