Answers From the Folders Already Sitting in Box
Box is where a lot of organizations keep contracts, reports, recordings and shared project folders, and that material usually sits outside whatever people search when they need an answer. Box publishes its own MCP server, and AI Intelligence Hub connects to MCP servers as a client, so an agent works against Box content with configuration rather than development. Box stays the place people file into.
How it connects
Box runs a hosted MCP server, and the platform is the client. The connection is configured once with a name, the server URL, the transport, and the HTTP headers that carry its credential. Streamable HTTP, Server-Sent Events and WebSocket are the transports carried, which is what reaching a vendor's hosted server needs. The server's tool list is fetched when the connection is bound rather than coded in, so a tool Box adds later is usable without reconfiguration.
Both forms are available. In a workflow, an MCP node runs one named tool with its arguments fixed at design time. For an agent, an MCP tool hands over the whole tool list and the agent chooses the tool and fills the arguments as it works. Where a step needs something the server does not expose, the HTTP Request node calls Box's own API.
The connection carries the credential it is configured with, so what an agent can read in Box is bounded by what that credential is granted there. That is worth settling at setup, because it is the boundary rather than a per-person one. Where an answer also draws on the platform's own library, that retrieval runs under the asking user's identity with their access list applied as a pre-filter, so library material a person could not open is never a candidate.
Bringing Box folders into the library on a schedule is a separate path and is built per engagement. It would run on the same ingestion connector architecture that already carries Google Drive, OneDrive and SharePoint, with an extension allowlist, excluded folders, the source hierarchy preserved or flattened, and the source file left, moved or deleted after each run.
What you can do together
- Ask AI Intelligence Hub a question that spans Box content and the recordings and documents already in the library, and get one answer with citations rather than two searches.
- Put a workflow on a Box folder so a document that lands there is read, classified and routed, with a person approving anything that writes back or sends.
- Reach Box's own search and its document tools through the server's tool list, without waiting for a connector to be built for each one.
- Have the folders you name ingested into the library instead, where the material needs transcription, retention and a custody record of its own. That path is scoped per engagement.
A scenario
- SetupAn administrator adds Box's MCP server to the portal once: name, URL, transport, and a credential scoped to the two project libraries. The tool list appears without anything being coded.
- Tuesday, 09:20A project manager asks AI Intelligence Hub which foundation option the March design review settled on. The agent reads the Box folders through the server and the recorded review from the library, and answers citing the drawing and the moment in the recording.
- WednesdayA workflow runs each morning over one Box folder, reads what arrived overnight, and drafts a summary for the project lead. A human-in-the-loop node holds it until the lead approves.
- LaterThe firm decides the closed-project archive needs retention and a custody record of its own, so the recorded reviews are scoped into an ingestion connector built for that purpose, and the working documents stay in Box.
What stays where
Box remains the system people file into
Folders, sharing and the day-to-day working copy stay in Box. The connection reads through Box's own server and writes only what a tool call is authorized to write.
The credential is the boundary on the Box side
What the agent can reach is what the configured credential is granted, so scoping it is the control. The platform's own library stays under its own access model, and retrieval from it runs as the asking user.
Nothing is replaced
People go on saving to Box exactly as they do now, and no second copy is created unless an ingestion connector is built to make one.
Processing runs where you deploy the platform
Shared or dedicated SaaS, your own cloud subscription, or your own servers. Reaching Box's hosted server needs a network path to it, which an air-gapped deployment does not have.
Products and solutions
Next step
See it on your own Box instance. We will show the connection made, the data moving and the output, then size it for your deployment.
Request a demonstration or write to sales@vidizmo.ai
sales@vidizmo.ai · vidizmo.ai/integrations/catalog/box