Claude Cowork OneDrive Connector: File Access & Benchmarks

When Claude Cowork crawls a user's Microsoft OneDrive, the native connector interrogates the Microsoft Graph /me/drive/root hierarchy recursively, turning directory listings and unindexed document payloads into active prompt tokens before synthesizing an answer.

The Claude Cowork OneDrive connector links Anthropic Claude directly to Microsoft OneDrive, enabling agents to access, read, and reference Microsoft 365 files during collaborative work.

Knowledge workers store millions of working documents across Microsoft OneDrive, Dropbox, Google Drive, Box, and SharePoint. In typical corporate environments, OneDrive serves as an employee's personal primary drive: a mix of active spreadsheet drafts, slide decks, private notes, and desktop backups. When teams deploy Claude Cowork as an autonomous agentic workspace, the immediate goal is connecting Claude to these personal and departmental files so the model can analyze datasets, prepare briefing memos, and draft project deliverables.

Anthropic addresses this requirement through a native Microsoft 365 connector. Under the hood, this integration connects to Microsoft Graph API endpoints using registered enterprise applications in Microsoft Entra ID. While this direct link allows Claude Cowork to search across an authenticated user's cloud drive, the underlying retrieval mechanism introduces severe operational frictions for high-volume agent workflows.

The native connector pulls entire folder structures rather than querying pre-indexed document chunks. When an agent searches for a specific figure or clause, it cannot simply fetch relevant text snippets from a pre-computed vector index. Instead, Claude Cowork must first traverse the directory hierarchy over Microsoft Graph. The model issues repetitive directory enumeration requests, inspecting folder items, subfolder hierarchies, and file metadata.

In an enterprise OneDrive account, directory trees are frequently broad and disorganized. OneDrive sync clients automatically redirect local folders using Known Folder Move, populating root directories with Desktop shortcuts, temporary application files, and hundreds of nested subfolders. When Claude Cowork crawls this unindexed structure, directory listings consume substantial context window capacity before the model ever reads document content.

Once the connector locates target documents, it downloads entire file payloads across the network into the model's active prompt context. If a user asks Claude to summarize key takeaways across several multi-megabyte Word documents, complex Excel workbooks, or large PDF slide decks, the agent ingests the entire text stream. Thousands of tokens get consumed by boilerplate headers, raw numerical tables, and irrelevant background narrative. This unindexed traversal quickly inflates token consumption, introduces network latency with each Graph API request, and increases the danger of truncated model outputs.

• How to Configure Devin AI Docker File Access and DevcontainersContainer usage among IT professionals reached 92% in 2025, highlighting the need for standard environments in software...
• How to Use OpenClaw File VersioningOpenClaw file versioning tracks changes to files in Fastio workspaces. It creates automatic snapshots on edits, shows...
• How to Connect Claude Code to OneDrive with Remote MCPClaude Code OneDrive integration connects Anthropic's terminal coding agent to Microsoft OneDrive folders, enabling CLI...
• Claude Cowork Box Connector: Connecting Team Storage & WorkspacesA Claude Cowork Box connector links Anthropic's collaborative agent workspace to enterprise Box repositories, allowing...
• Claude Cowork Dropbox: Native Connector Setup and Retrieval PerformanceThe Claude Cowork Dropbox connector allows Anthropic Claude to directly browse, search, and ingest files stored in...
• Claude Cowork SharePoint: Connecting Document Libraries via MCPConnecting Claude Cowork to SharePoint document libraries allows AI agents to query enterprise knowledge without broad...

More on this subject: Claude and Claude Code (249 guides)

Setting up Claude Cowork with OneDrive through native tooling requires navigating administrative security controls in Microsoft 365. The native integration operates through two registered enterprise applications in Microsoft Entra ID: M365 MCP Client for Claude and M365 MCP Server for Claude.

Activating these applications requires an organization's Microsoft Entra Global Administrator to grant tenant-wide administrative consent. This approval authorizes broad delegated permission scopes, including User.Read, Files.Read, and Files.Read.All. If agents need to write outputs back to OneDrive, administrators must consent to write scopes such as Files.ReadWrite.All.

For IT security teams, granting tenant-wide administrative consent creates an uncomfortably broad blast radius. In a corporate Microsoft 365 tenant, an employee's OneDrive contains private working documents: personal expense filings, draft compensation proposals, preliminary performance reviews, and sensitive project notes. Because the native connector uses broad delegated user credentials to query the Graph API, it cannot enforce folder-specific access boundaries. Any file readable by that employee can be returned in search results, creating data exposure risks if internal permissions are loose.

Account tiering introduces an immediate roadblock for many users. As documented in the Claude Help Center: "The Microsoft 365 connector requires a Microsoft Entra tenant tied to a Microsoft Business plan. Personal Microsoft accounts (such as @outlook.com or @hotmail.com addresses) can't be used to connect." Independent contractors, consultants, and teams relying on consumer Microsoft accounts or standalone personal OneDrive storage cannot enable the native connector.

Enterprise Conditional Access policies also create persistent friction for cloud agent runtimes. Anthropic executes Claude Cowork server requests from a dedicated IP address block (160.79.104.0/21). Corporate security baselines that enforce device compliance, require trusted office IP locations, or enforce short re-authentication intervals frequently block these incoming requests. IT teams encounter Entra errors such as AADSTS53003 (blocked by conditional access policy) or AADSTS70043 (sign-in frequency expired), necessitating custom named IP locations and policy exclusions in the Entra admin center.

File format compatibility adds further constraints. While the native connector reads standard office formats (.docx, .xlsx, .pptx), text files (.txt, .md, .csv), and PDFs, it cannot parse proprietary formats, and Microsoft OneNote notebooks (application/msonenote) return MIME type errors when requested.

Third-party integration services offer an alternative routing approach. As noted in documentation from Composio: "With a standalone One drive MCP server, the agents and LLMs can only access a fixed set of One drive tools tied to that server. However, with the Composio Tool Router, agents can dynamically load tools from One drive and many other apps based on the task at hand, all through a single MCP endpoint." While Composio manages OAuth credentials and dynamic tool discovery, the agent still executes live requests against Microsoft Graph endpoints, leaving the underlying performance challenges of unindexed file downloads unresolved.

Rather than providing Claude Cowork full-tenant credentials that traverse unconstrained cloud accounts, teams can isolate document assets by connecting dedicated OneDrive folders to a Fastio workspace. This architecture maintains OneDrive as the familiar filing surface for business personnel while providing AI agents with an isolated retrieval boundary.

Fastio Cloud Sync establishes scheduled or on-demand file transfers between Microsoft OneDrive and Fastio workspaces (with two-way write-backs or one-way mirroring). Sync routines run in predictable batches rather than continuous live mirroring, preserving system performance. (Cloud Sync also supports Dropbox and Box, while Google Drive operates via cloud import with synchronization in development; SharePoint document libraries connect via the OneDrive integration). By isolating a single project directory, such as an audit repository or acquisition folder, organizations ensure Claude Cowork never touches unrelated personal files, executive correspondence, or personal desktop backups housed elsewhere on OneDrive.

Within the workspace, Fastio's Intelligence Mode activates automated retrieval indexing. The engine ingests office documents, PDF manuals, spreadsheets, and markdown notes as soon as sync transfers finish. Instead of requiring developers to build custom retrieval pipelines, host vector databases, or calculate text chunk embeddings, Fastio automatically generates hybrid search indices combining dense semantic vectors, BM25 keyword matching, and document metadata. When Claude Cowork requests information, the workspace delivers relevant text excerpts with exact source references, preventing large unindexed files from clogging the agent's context window.

Teams managing high volumes of semi-structured business files can also implement Metadata Views. Metadata Views convert unstructured document collections into structured, spreadsheet-like databases. By defining field definitions in plain language, users direct Fastio to extract typed attributes (such as dates, invoice totals, vendor entities, jurisdiction flags, and URLs) across synchronized OneDrive files. The extraction engine processes documents, spreadsheets, and scanned PDFs without requiring rigid OCR configurations. Claude Cowork can inspect these populated views or request custom schema extractions over MCP, granting agents instant access to clean tabular data without downloading multi-page files.

Connect OneDrive to Claude Cowork with pre-indexed retrieval

Sync OneDrive folders into a persistent Fastio workspace, auto-index content for hybrid semantic search, and query via remote MCP. Monthly plans start with a 30-day free trial.

Start 30-Day Trial

Connecting Claude Cowork to OneDrive through Fastio involves setting up scoped folder sync, enabling workspace intelligence, and registering Fastio's remote Model Context Protocol endpoint in Claude's configuration.

1. Configure Scoped Folder Sync from OneDrive

Begin by creating an isolated workspace and linking your target OneDrive folder:

1. Sign in to your Fastio organization dashboard and create a workspace for the project (such as financial-due-diligence or product-roadmaps).
2. In the workspace navigation bar, open Workspace Settings and select Cloud Sync.
3. Select the OneDrive integration and complete OAuth authorization with your Microsoft 365 credentials.
4. Select the specific OneDrive folder or subfolder containing the project documents.
5. Select the sync direction: choose one-way to mirror OneDrive files into Fastio, or two-way to allow agent deliverables created in Fastio to sync back to OneDrive.
6. Set the sync schedule (such as hourly, daily, or on-demand).

Files transfer directly between cloud providers without consuming local workstation storage or hydrating Files On-Demand stubs on local machines.

2. Turn On Workspace Intelligence Mode

With files populated in the workspace, activate automated document indexing:

1. In Workspace Settings, click Intelligence Mode.
2. Switch Intelligence to enabled. Fastio immediately indexes all supported documents, generating semantic vector embeddings and keyword search indexes.
3. Monitor indexing status in the workspace overview. Once complete, every document in the workspace is accessible via hybrid semantic search.

3. Add Fastio Remote MCP to Claude Cowork

Fastio hosts a remote Model Context Protocol server at https://mcp.fast.io/mcp/tools over Streamable HTTP.

To connect Claude Cowork or Claude Desktop, open Customize, then Connectors; add a custom connector and paste https://mcp.fast.io/mcp/tools; select Connect and sign in to Fastio in the window that opens; in a chat, turn Fastio on from the + menu under Connectors. Sign-in shows a Review Permissions screen where you pick Read Only or Read & Write and which organizations and workspaces the connection can reach.

Because Fastio provides a hosted remote MCP server, no local Node.js packages, npx launchers, or stdio child processes run on your machine. Claude Cowork connects directly to the cloud endpoint over HTTP. For complete configuration recipes, review the setup documentation or inspect agent onboarding guidance at https://fast.io/llms.txt.

4. Query Documents via Hybrid Semantic Search

Once connected, Claude Cowork uses Fastio's consolidated MCP toolset to query documents. Rather than downloading entire files over Microsoft Graph, Claude executes hybrid semantic search queries against GET /current/workspace/{workspace_id}/storage/search/:

{
  "name": "storage",
  "arguments": {
    "action": "search",
    "workspace_id": "ws_7c9d2e1f",
    "search": "capital expenditure projections and operating cash flow",
    "folders_scope": ["fld_quarterly_briefs"]
  }
}

The Fastio workspace evaluates the query using hybrid semantic search and returns targeted document excerpts with exact filenames and page references. Claude Cowork synthesizes the final response from these verified passages, keeping prompt context clean and token consumption minimal.

5. Multi-Agent Safeguards and Human Handoff

When autonomous agents collaborate with human teammates across synchronized OneDrive folders, Fastio provides foundational operational protections:

• Per-File Version Tracking: Every file write automatically creates an independent version entry. If an agent revises an executive report or generates an updated forecast, previous versions remain immediately recoverable through the dashboard or API.
• Advisory File Leases: Agents can request advisory locks using the MCP storage_manage tool (lock-acquire, lock-status, and lock-release). These non-exclusive leases inform teammates and other agents that editing is in progress without hard-locking storage, allowing concurrent changes to merge cleanly into version history.
• Administrative Ownership Transfer: An automated agent can establish an organization, set up project workspaces, and configure sync connections before transferring primary ownership to a human team lead while retaining operational administrator credentials.
• Immutable Audit Trail: All file views, modifications, permissions adjustments, and sync events are recorded in an append-only audit log for operational traceability.
• Activity Feed Subscriptions: External agents and services can follow workspace activity using WebSocket connections or the HTTP long-poll route (GET /current/activity/poll/{entity_id}?wait=95&lastactivity={timestamp}) to trigger subsequent actions when new files sync from OneDrive.

Evaluating direct Microsoft Graph directory traversal against pre-indexed workspace retrieval demonstrates fundamental differences in operational efficiency. When Claude Cowork queries OneDrive directly through native connectors, latency compounds with folder depth and file size. The agent repeatedly executes directory listing tools, filters metadata client-side, and pulls raw document streams into its context window. In contrast, Fastio offloads document parsing and search execution to the workspace server. Claude receives pre-filtered, highly relevant passages that fit cleanly into prompt context.

Independent performance measurements demonstrate the efficiency gains of remote indexed retrieval compared to native cloud storage connectors. The connector comparison is published at https://fast.io/benchmarks/ and that page is the only place its numbers live. In that published evaluation, Fastio was measured the fastest and the lowest cost of the providers tested. Every measured row represents that provider's native connector in Claude Cowork; no row represents a local stdio server, a Files-On-Demand stub, or SharePoint, and SharePoint was not measured at all. Accuracy was parity across tested providers, and per-run scores are not published.

Organizations evaluating persistent workspaces can choose between three subscription tiers, starting with a 14-day free trial on monthly plans that requires a credit card:

Metered usage beyond included tier allocations follows standard published rates:

External guest recipients opening shared links do not consume organization member seats, keeping overhead stable as teams scale collaborative agent operations.