SharePoint MCP Server: Secure Setup for Claude & AI Agents (2026)
The SharePoint MCP server lets Claude, Cursor, ChatGPT, and AI agents read and act inside SharePoint. Here's the official setup, the real security risks, and how to deploy it with DLP-grade redaction at the MCP layer.
The SharePoint MCP server is the path for AI agents (Claude, Cursor, ChatGPT, Perplexity, custom agents) to read and act inside SharePoint via the Model Context Protocol — covering SharePoint sites, document libraries, pages, lists, and collaboration metadata.
Setup is documented in the official SharePoint MCP server guide; connecting from Claude Desktop requires the Enterprise/Pro/Max/Team plan plus an OAuth client ID/secret added as a custom connector.
The risk: every SharePoint MCP tool call returns the data the authorizing user can see. That data routinely contains PII, PHI, financial records, contracts, source code, secrets, and credentials. None of it is inspected before reaching the AI model's context window.
Strac SharePoint MCP DLP governs every tool call between the AI agent and SharePoint. Strac controls what each agent can reach and do across your sites and document libraries (allow, block, or require approval on high-risk actions), protects the content it returns — including text inside images (redact, mask, vault, OCR) — and logs every call as audit evidence mapped to SOC 2 / HIPAA / PCI / GDPR / EU AI Act / ISO 42001. One control plane: see, control, protect, prove.
Setup is agentless and under 10 minutes per workspace. No application code changes, no agent SDK changes, no SharePoint re-permissioning.
What Is the SharePoint MCP Server?
The SharePoint MCP server is a Model Context Protocol implementation that exposes SharePoint's API as a standardized set of tools to AI agents. Once connected, an agent like Claude can perform SharePoint search, site get, document fetch, list query, collaborations get on the authenticated user's behalf — turning SharePoint's API surface into AI-actionable capabilities.
Refer to the official SharePoint MCP server documentation for the current tool list, OAuth scopes, and rate-limit behavior. The setup pattern is consistent with other MCP integrations: an OAuth client ID/secret, a custom connector in Claude (or another MCP-aware AI client), and the server starts serving tool calls.
From the user's perspective, the AI agent suddenly knows their SharePoint. From the security perspective, the AI agent now has read access — and often write access — to every record the user can touch in SharePoint.
That's the value. It's also where security teams need a control layer.
What AI Agents Can Actually Do With SharePoint MCP
Put the security question aside for a moment; the pull is pure productivity. Wired into Claude, Cowork, or another MCP client, the SharePoint MCP server lets an agent run real work against your tenant:
Hunt across sites and document libraries — point the agent at a question ("where's the renewed vendor agreement?") and it queries across site collections and libraries instead of you clicking through nested folders.
Read and summarize documents and list items — the agent opens a proposal deck, a project list row, or a meeting-notes page and hands back a tight summary, key dates, or a decision log without anyone downloading the file.
Pull text out of images and scanned files — OCR runs over embedded screenshots, scanned PDFs, and image attachments, so a photographed contract or a scanned ID becomes searchable, quotable text the agent can reason over.
Gather policy, contract, and HR documents spread across collections — the agent assembles the current PTO policy, an MSA, and an onboarding checklist that live on three different team sites into one answer.
Surface what changed recently — ask "what got touched in the deal room this week?" and the agent returns the most recently modified files and list entries, ranked by edit time.
Draft and write back into SharePoint — generate a status page, append a row to a tracking list, or update document metadata, all initiated from the agent.
That reach is exactly why each agent's access and actions need to be controlled across sites and libraries, the document data it pulls back protected, and every one of those tool calls audited — governance over the full SharePoint MCP path, not inspection alone.
The Real Security Risks of the SharePoint MCP Server
The risks fall into four categories that every healthcare, fintech, and enterprise security team should price into the deployment.
1. SharePoint search walks the user's full read scope.sharepoint_search_documents traverses every site the authorising user can read — including inherited sites from prior roles, diligence rooms long-since closed, and shared sites the user joined years ago and forgot.
2. Document content is PII-, PHI-, and contract-dense by default.sharepoint_get_document returns the raw bytes — PDFs and Office documents routinely contain customer records, salary data, M&A diligence material, contracts, signed agreements, and credentials pasted into operational runbooks.
3. Public links and external guests multiply the exposure.sharepoint_list_collaborations returns every external sharing relationship per site — publicly accessible links, guest users, anonymous links. AI agents reading SharePoint inherit all of it. Many tenants do not know what is publicly exposed until Strac’s discovery scan completes.
4. Web parts and custom lists leak regulated metadata. Custom SharePoint lists and web parts hold customer IDs, account numbers, internal classifications, and personnel data — data classes traditional file DLP rarely inspects. MCP tool calls return list rows by default.
The traditional DLP a company already runs — at the network edge, on the file share, inside the SaaS-native rule engine — does not sit in the MCP path. The tool response goes straight from SharePoint into the AI agent's context window. That gap is where Strac SharePoint MCP DLP lives.
Strac's SharePoint MCP DLP is the governance layer between AI agents and the SharePoint MCP server. It gives you four things on every tool call: see which agent touched which site, library, or list; control what each agent is allowed to reach and do — allow, block, or require approval on high-risk actions like writes and external shares; protect the content that flows back, redacting, masking, or vaulting sensitive data per policy; and prove it all with a per-call audit record. Strac intercepts every tool call before content reaches the AI agent's context window, and non-sensitive content flows through untouched.
The Strac SharePoint MCP DLP gateway intercepts every tool call between any AI agent (Claude, Cursor, Cowork, ChatGPT, custom) and the SharePoint MCP server. PII, PHI, PCI, secrets, source code, and content inside images are redacted before the AI agent ever reads them.The full data flow: a user prompt triggers an AI agent tool call, the MCP server fetches from SharePoint, and the Strac DLP redaction engine strips SSNs, credit cards, emails, PHI, secrets, and source code before the redacted response ever reaches the model.Strac's live MCP Access console — every AI agent tool call touching SharePoint and your other connected platforms, captured and inspected for sensitive data in real time. See what your LLMs reached for, who prompted, and what was flagged.Every MCP invocation in order — user, tool, platform, and the sensitive data found — with redacted vs. original content and a full audit trail. This is what Strac shows on SharePoint that access-only gateways can't: the data in each call, not just the call.
Why not just an access gateway?
Access-only tools answer "who called what." They do not see the regulated data in a synced document library. Strac sits inline on every SharePoint tool call: it detects and remediates the sensitive data inside — redact, mask, block, or revoke access — and approves or blocks risky actions per agent, and keeps the audit trail — the call and its contents.
What Strac does on every SharePoint tool call
One inline pass over each MCP response — five actions, enforced by your policy:
Detect — finds regulated data in a library and any PII, PHI, PCI, secrets, or source code in the payload, including text inside images via OCR.
Redact or mask — replaces the sensitive elements inline, so the agent still gets its answer and the model never sees the raw data.
Block or require approval — stops a high-risk action like an external-member grant, or routes it for sign-off before it runs.
Alert — notifies your team and streams the event to your SIEM (Microsoft Sentinel, Datadog, Splunk) in real time.
Audit — logs who, which agent, which tool, what data, and the action taken — evidence mapped to PCI DSS, SOC 2, HIPAA, and GDPR.
What this looks like in practice:
Read tools are filtered. When the agent calls a read tool, Strac inspects the returned payload, redacts SSNs / credit cards / emails / PHI / API keys / secrets / source code inline, and passes the clean payload to the agent. The agent still does its job; the regulated data never enters the model context.
Write tools are guardrailed. When the agent invokes a write/post/create tool with content that contains sensitive data, Strac inspects the outgoing payload and either redacts, vaults, or blocks depending on the channel and the data type.
Files, attachments, images, and documents are inspected at depth. PDFs, DOCX, XLSX, ZIPs, and image attachments are parsed with the same OCR and document-parser pipeline Strac uses across its DLP product line. Sensitive content inside screenshots and scanned PDFs is found and redacted.
Every invocation is logged. AI client, user, tool name, resource accessed, data classes detected, redactions applied, vault references, disposition. The log is the SOC 2 / HIPAA / PCI / GDPR audit evidence — produced automatically.
Policy is contextual. Different resources, different policies. Strac maps to your existing data classification, not an MCP-specific silo.
The same Strac MCP DLP layer covers Claude Cowork, Slack MCP, and other surfaces — one control plane across every place AI agents touch your regulated data.
✨ Strac Native SharePoint DLP — The Companion to MCP DLP
MCP DLP protects the AI-agent surface. Strac's native SharePoint DLP protects the direct-user surface — the same SharePoint workspace, but inspected at the point where humans share, upload, send, and grant access. Most enterprises run both: native DLP for the user-driven actions, MCP DLP for the agent-driven actions. Together they cover every path regulated data can take in and out of SharePoint.
Strac SharePoint DLP — data classification, labelling, and remediation policy in action
What Strac's native SharePoint DLP includes:
Continuous discovery and classification of PII, PHI, PCI, credentials, contracts, and source code across every SharePoint site and document library — the most-deployed enterprise content surface in 2026
Real-time scanning of every uploaded document — DOCX, PDF, XLSX, PPTX, ZIPs — plus OCR on images and image-based PDFs (W-2s, IDs, scanned contracts)
Discovery of publicly-shared SharePoint links and externally-shared documents across the tenant — a finding that often surfaces years of accumulated exposure on day one
Discovery of external-tenant guests with access to SharePoint sites — diligence-room collaborators, vendors, contractors, and consultants whose access outlived the engagement
Automatic remediation: revoke public links, remove external members, restrict over-permissive permissions, label-and-redact regulated content — configurable per data class
Vault-redaction flow: replace sensitive documents with permission-controlled retrieval links so the content stays accessible to authorised users without sitting in SharePoint in cleartext
Honours your Microsoft Purview classification labels — an item marked Highly Confidential in SharePoint stays Highly Confidential when an AI agent retrieves it
Audit logs per finding mapped to SOC 2 CC6, HIPAA Security Rule, PCI Req. 3/4/7/10, ISO 27001 A.5.12/A.5.13/A.8.10-12, and GDPR Art. 5/25/30/32
For the broader integration catalog — every SaaS, cloud, browser, and endpoint surface Strac covers — see strac.io/integrations.
✨ See Strac MCP DLP in Action
The screenshot below shows Strac's MCP DLP redacting sensitive data from a real Claude session — patient identifiers, customer emails, and credit card numbers tokenized inline before the model received the prompt. The same inspection pattern runs on every SharePoint MCP tool call routed through Strac.
Strac DLP at work inside a Claude conversation: sensitive elements tokenized inline before the model sees them. The same pattern runs at the MCP layer for every SharePoint tool call.
How to Set Up Strac SharePoint MCP DLP
Setup is agentless and takes under 10 minutes.
Authorize Strac with your SharePoint tenant via OAuth. Strac requests the read/write scopes for the products you want covered. Honors SharePoint's permission model — Strac only sees what the authorizing user/bot can see.
Configure the MCP proxy endpoint. Strac issues an MCP server endpoint that drops into your AI client's MCP configuration. For Claude Desktop:
json
"mcpServers": {
"sharepoint": {
"url": "https://mcp.strac.io/sharepoint",
"auth": { "type": "bearer", "token": "<your-strac-token>" }
}
}
For Cursor, OpenAI Agents, custom agents — same endpoint, same auth.
Pick your policy. Out-of-the-box templates for SOC 2, HIPAA, PCI, GDPR. Custom policies (resource-level, data-class-level, action-level) take minutes to configure.
Done. Every MCP tool call between your agent and SharePoint now flows through Strac. No application code changes. No agent code changes. The audit log starts populating immediately.
Compliance Coverage Out of the Box
The same Strac SharePoint MCP DLP control produces evidence mapped to every major compliance framework.
Framework
What Strac SharePoint MCP DLP Satisfies
SOC 2
CC6.6 (unauthorized data exposure), CC6.7 (restricted transmission of data to external systems), CC7.2 (monitoring for anomalies including AI usage)
The SharePoint MCP server is a Model Context Protocol implementation that lets AI agents (Claude, Cursor, ChatGPT, Perplexity, custom agents) read and act inside SharePoint via standardized tool calls. It's how an AI assistant gets contextual access to SharePoint sites, document libraries, pages, lists, and collaboration metadata.
Is the SharePoint MCP connector the same as the SharePoint MCP server?
Two names, one component. The MCP spec says server; in Claude and Cursor it appears as the SharePoint connector. Both give an agent the same reach across sites and document libraries, and Strac's SharePoint MCP connector redacts content — including text inside images — before the model sees it.
SharePoint MCP vs Microsoft 365 Copilot — what's the difference?
They sit on opposite sides of the boundary. The SharePoint MCP server is how an external agent — Claude, Cursor, or any MCP-aware client running outside Microsoft's stack — reaches into SharePoint over the Model Context Protocol to read and write content. Copilot for SharePoint and the broader M365 Copilot are Microsoft's native, in-product AI: the assistant lives inside the tenant, grounded on Microsoft Graph, and answers within the Microsoft surface. The practical distinction is where the data lands — Copilot keeps content inside Microsoft's perimeter, while the MCP server hands SharePoint content back out to a third-party AI client. That hand-off on the tool-call return path is precisely where Strac SharePoint MCP DLP inspects each response and redacts PII, PHI, PCI, secrets, and source code before it reaches the external model.
Is the SharePoint MCP server safe to use with sensitive data?
By itself, no — not without an additional DLP layer. The SharePoint MCP server honors the authorizing user's permissions but returns whatever that user can see, including PII, PHI, credentials, source code, and other regulated content. For enterprise use with regulated data, you need an MCP-layer DLP control like Strac SharePoint MCP DLP that inspects and redacts every tool response before content reaches the AI model.
How is Strac SharePoint MCP DLP different from SharePoint's built-in protections?
SharePoint's built-in protections operate at the storage and policy layer — sensitivity labels, retention policies, native DLP rules at posting/sharing time. None of those sit in the MCP tool-call path by default. Strac is purpose-built for the MCP layer: it inspects every tool response before content reaches the AI agent's context window, with detection breadth (PII / PHI / PCI / secrets / source code / OCR-in-images) that goes well beyond most native rule engines.
Does Strac SharePoint MCP DLP work with Claude, Cursor, ChatGPT, Cowork, and custom agents?
Yes. Strac exposes a standard MCP endpoint, so any MCP-aware AI client routes tool calls through it with one configuration change. No SDK changes, no application code changes.
What sensitive data types does Strac detect in SharePoint MCP tool responses?
PII (SSN, driver's license, passport, address, phone, email), PHI (clinical notes, MRN co-occurrence, ICD-10 codes adjacent to identifiers, lab values), PCI (full and partial card numbers via Luhn check), credentials (API keys, AWS / GCP / Azure access keys, OAuth tokens, JWTs, SSH keys, private keys — 48+ patterns), proprietary content (M&A keywords, source code fingerprints), and custom detectors trained on your internal data classifications. Detection runs across text, files, images (OCR), and structured fields.
How long does Strac SharePoint MCP DLP take to deploy?
Under 10 minutes for the first workspace. OAuth Strac into SharePoint, paste the Strac MCP endpoint into your AI client's config, pick a policy template, done. No agents to install, no SharePoint re-permissioning, no application code changes.
Where does redacted data go — is it stored?
Redacted content is replaced inline in the tool response. Optionally, sensitive content can be vaulted — replaced with a short-lived retrieval link that only authorized users can resolve, so the original data is retrievable for legitimate use without ever entering the AI context. Vaulted data is stored encrypted at rest in your Strac tenant; you control retention.
Can I see what an AI agent did in my SharePoint workspace?
Yes. Strac produces a per-call audit log: timestamp, AI client identity, user, tool invoked, resource accessed, data classes detected, redactions applied, vault references, disposition. The log is queryable in the Strac console and exportable to your SIEM. This is the evidence trail SOC 2, HIPAA, PCI, and GDPR auditors will ask about for AI-agent activity in SharePoint.
The Bottom Line
The SharePoint MCP server is rapidly becoming the way AI agents read into SharePoint. That surface contains every category of regulated and proprietary data your organization has. Running SharePoint MCP in 2026 without an MCP-layer DLP control is not a question of if the first incident reaches your security team; it's when.
Strac SharePoint MCP DLP gives you the protection layer, the audit evidence, and the framework-agnostic compliance coverage so you can let your team use SharePoint with Claude, Cursor, Cowork, ChatGPT, and any future AI client without making each one a separate security exception.
If you are running — or about to run — SharePoint MCP in production, book a 30-minute demo. We'll walk through the architecture, the policy templates, and a deployment plan for your specific SharePoint workspace and AI clients.
The SharePoint MCP server is a Model Context Protocol implementation that lets AI agents (Claude, Cursor, ChatGPT, Perplexity, custom agents) read and act inside SharePoint via standardized tool calls. It's how an AI assistant gets contextual access to SharePoint sites, document libraries, pages, lists, and collaboration metadata.
Is the SharePoint MCP connector the same as the SharePoint MCP server?
Two names, one component. The MCP spec says server; in Claude and Cursor it appears as the SharePoint connector. Both give an agent the same reach across sites and document libraries, and Strac's SharePoint MCP connector redacts content — including text inside images — before the model sees it.
SharePoint MCP vs Microsoft 365 Copilot — what's the difference?
They sit on opposite sides of the boundary. The SharePoint MCP server is how an external agent — Claude, Cursor, or any MCP-aware client running outside Microsoft's stack — reaches into SharePoint over the Model Context Protocol to read and write content. Copilot for SharePoint and the broader M365 Copilot are Microsoft's native, in-product AI: the assistant lives inside the tenant, grounded on Microsoft Graph, and answers within the Microsoft surface. The practical distinction is where the data lands — Copilot keeps content inside Microsoft's perimeter, while the MCP server hands SharePoint content back out to a third-party AI client. That hand-off on the tool-call return path is precisely where Strac SharePoint MCP DLP inspects each response and redacts PII, PHI, PCI, secrets, and source code before it reaches the external model.
Is the SharePoint MCP server safe to use with sensitive data?
By itself, no — not without an additional DLP layer. The SharePoint MCP server honors the authorizing user's permissions but returns whatever that user can see, including PII, PHI, credentials, source code, and other regulated content. For enterprise use with regulated data, you need an MCP-layer DLP control like Strac SharePoint MCP DLP that inspects and redacts every tool response before content reaches the AI model.
How is Strac SharePoint MCP DLP different from SharePoint's built-in protections?
SharePoint's built-in protections operate at the storage and policy layer — sensitivity labels, retention policies, native DLP rules at posting/sharing time. None of those sit in the MCP tool-call path by default. Strac is purpose-built for the MCP layer: it inspects every tool response before content reaches the AI agent's context window, with detection breadth (PII / PHI / PCI / secrets / source code / OCR-in-images) that goes well beyond most native rule engines.
Discover & Protect Data on SaaS, Cloud, Generative AI
Strac provides end-to-end data loss prevention for all SaaS and Cloud apps. Integrate in under 10 minutes and experience the benefits of live DLP scanning, live redaction, and a fortified SaaS environment.