Overview

This MCP (Model Context Protocol) server provides a simple bridge between LLMs and Microsoft 365 file storage (SharePoint and OneDrive). It lets developers expose a searchable, protocol-compatible tool that returns matching documents (PDF, DOCX), spreadsheets, presentations, and common image formats. The server is designed to be used as a tool endpoint for LLMs that support MCP or tool-calling, enabling retrieval-augmented workflows and contextual document lookup.

Why this is useful: instead of manually crawling or building your own SharePoint/OneDrive integration, this server handles authentication against Microsoft Graph, file-type filtering, and a lightweight search API that returns references and content snippets (or file downloads). It’s optimized for developer integration into conversational agents, knowledge workers, and automated document processing pipelines.

Features

Search SharePoint and OneDrive using Microsoft Graph search APIs
Return results for PDFs, DOCX, PPTX, XLSX and common image formats
MCP-compatible tool descriptions for use with tool-aware LLMs
Pagination and simple ranking / snippet extraction
Authentication via Azure app registration (OAuth client credentials)
Lightweight JSON API suitable for wrapping in chains or tool calls

Installation / Configuration

Prerequisites:

Node.js (14+)
An Azure AD application with Microsoft Graph permissions to access SharePoint and OneDrive (Files.Read.All or Sites.Read.All)

Clone and install dependencies

git clone https://github.com/godwin3737/mcp-server-microsoft365-filesearch.git
cd mcp-server-microsoft365-filesearch
npm install

Configure environment variables

Create a .env file in the project root (or set env vars in your host). Required variables:

PORT=8080
AZURE_TENANT_ID=your-tenant-id
AZURE_CLIENT_ID=your-client-id
AZURE_CLIENT_SECRET=your-client-secret
ALLOWED_SITE_IDS=site-id-1,site-id-2       # optional: comma-separated SharePoint site IDs to restrict search
MAX_RESULTS=50                            # optional: limit results per request

Table: Key environment variables

Variable	Description
AZURE_TENANT_ID	Azure AD tenant ID for your Microsoft 365 organization
AZURE_CLIENT_ID	Azure AD application (client) ID
AZURE_CLIENT_SECRET	Client secret for the Azure AD app
ALLOWED_SITE_IDS	Optional comma-separated site IDs to restrict search scope
PORT	Server port to listen on (default 8080)

Start the server

npm start
# or
node server.js

The server starts on the configured port and exposes MCP-compatible endpoints.

Available Tools / Resources

The server exposes a small set of HTTP endpoints and MCP tool descriptions to integrate with LLMs.

GET /mcp/tools
- Returns a JSON array of MCP tool descriptions (name, input schema, description).
- Typical tool: “microsoftFilesSearch” — accepts a text query and optional filters (fileType, siteId, top).
POST /search
- Payload: { “query”: “contract renewal”, “fileTypes”: [“pdf”,“docx”], “siteId”: “…” , “top”: 10 }
- Returns: list of results with metadata: { id, name, fileType, url, snippet, lastModified }
GET /download/:itemId
- Returns direct file stream for a given Microsoft Graph item id (requires auth and ACLs).

Example /search request:

POST /search
Content-Type: application/json

{
  "query": "yearly financial report",
  "fileTypes": ["xlsx", "pdf"],
  "top": 5
}

Example response (trimmed):

[
  {
    "id": "01ABC...",
    "name": "FY2025_financials.xlsx",
    "fileType": "xlsx",
    "url": "https://graph.microsoft.com/v1.0/sites/.../drive/items/01ABC",
    "snippet": "Summary: Revenue increased 12% year-over-year...",
    "lastModified": "2026-02-01T12:34:56Z"
  }
]

Use Cases

Conversational enterprise assistant
- Integrate the MCP tool into an LLM agent so users can ask “Show me the latest contract with Contoso” and the model calls the microsoftFilesSearch tool, retrieves matching DOCX/PDF files, and includes file snippets in the response.
Retrieval-augmented generation (RAG)
- Use the server to fetch relevant document contents (or snippets) for RAG pipelines. The agent queries /search, downloads top documents, and constructs a context window for the LLM to answer questions grounded in company documents.
Content discovery dashboard
- Build a UI that calls the /search endpoint to populate search results with previews, file type icons, and download links for users who need quick access to shared files across SharePoint and OneDrive.
Automated processing pipelines
- Use the API to locate newly uploaded or updated spreadsheets/presentations and trigger downstream ETL, analysis, or indexing workflows.

Notes and Best Practices

Permissions: Run with least privilege — grant only the Graph API permissions you need (prefer Sites.Read.All or Files.Read.All for read-only scenarios).
Rate limits: Microsoft Graph enforces throttling. Implement retries and backoff for large-scale search workloads.
Security: Host the server behind your application firewall and restrict access to trusted clients or internal networks. Validate or restrict ALLOWED_SITE_IDS to prevent broad data exposure.
Snippets & privacy: Snippet extraction can reveal content; ensure your usage complies with data handling and compliance requirements.

For developer reference, use the /mcp/tools endpoint to programmatically discover the tool schema and integrate the server into tool-enabled LLM frameworks.

MCP Server: SharePoint and OneDrive File Search