Microsoft MCP
Model Context Protocol servers for Microsoft 365, enabling AI assistants to interact with Calendar, Contacts, OneDrive, Outlook, and SharePoint via the Microsoft Graph API.
README
<div align="center">
<img src=".github/assets/icons/outlook.png" width="54" alt="Outlook" /> <img src=".github/assets/icons/calendar.png" width="54" alt="Calendar" /> <img src=".github/assets/icons/onedrive.png" width="54" alt="OneDrive" /> <img src=".github/assets/icons/sharepoint.png" width="54" alt="SharePoint" /> <img src=".github/assets/icons/contacts.png" width="54" alt="Contacts" />
<h1>Microsoft MCP</h1>
<p> <b>Model Context Protocol servers for Microsoft 365.</b><br/> Calendar · Contacts · OneDrive · Outlook · SharePoint — on the official <a href="https://www.npmjs.com/package/@modelcontextprotocol/sdk"><code>@modelcontextprotocol/sdk</code></a>, over stdio or Streamable HTTP. </p>
</div>
Each server speaks the real MCP protocol and runs over either transport:
- stdio — for local MCP clients that launch the server as a subprocess (Claude Desktop, IDEs, the MCP Inspector).
- Streamable HTTP — for remote/hosted use, with the Microsoft Graph access token supplied per request via
Authorization: Bearer.
Servers
| Server | npm package (and binary) | Tools | |
|---|---|---|---|
| <img src=".github/assets/icons/calendar.png" width="22" alt=""/> | Calendar | ms-calendar-mcp |
9 |
| <img src=".github/assets/icons/contacts.png" width="22" alt=""/> | Contacts | ms-contacts-mcp |
7 |
| <img src=".github/assets/icons/onedrive.png" width="22" alt=""/> | OneDrive | ms-onedrive-mcp |
9 |
| <img src=".github/assets/icons/outlook.png" width="22" alt=""/> | Outlook | microsoft-outlook-mcp |
14 |
| <img src=".github/assets/icons/sharepoint.png" width="22" alt=""/> | SharePoint | ms-sharepoint-mcp |
23 |
All tools are thin wrappers over the Microsoft Graph v1.0 API.
Layout
microsoft-mcp/
├── apps/ # one MCP server per Microsoft 365 product
│ ├── calendar/
│ ├── contacts/
│ ├── onedrive/
│ ├── outlook/
│ └── sharepoint/
│ └── src/
│ ├── tools.ts # declarative tool definitions (schema + handler)
│ └── index.ts # run({ name, version }, tools)
└── packages/ # shared building blocks
├── core/ # MCP server bootstrap + dual transport (stdio / HTTP)
├── graph/ # Microsoft Graph HTTP client
├── validation/ # id / path / query sanitizers
└── logger/ # structured JSON logging (stderr-only — stdio-safe)
A server is just a list of tools handed to run():
// apps/calendar/src/index.ts
import { run } from "@microsoft-mcp/core";
import { tools } from "./tools.js";
void run({ name: "microsoft-calendar", version: "1.0.0", title: "Microsoft Calendar" }, tools);
// a single tool
defineTool({
name: "get_event",
description: "Get a single calendar event by ID.",
inputSchema: { event_id: z.string().describe("Event ID") },
confirmationPolicy: "never",
handler: ({ graph }, { event_id }) => {
validateId(event_id, "event_id");
return graph.request("GET", `/me/events/${event_id}`);
},
});
confirmationPolicy ("always" for mutating/destructive tools, "never" for read-only) is surfaced to clients as MCP readOnlyHint / destructiveHint annotations.
Requirements
- Node.js >= 20
- pnpm 10 (
corepack enable)
Setup
pnpm install
pnpm build # build all servers (turbo) -> apps/*/dist/index.js
pnpm check-types # typecheck everything
Tests & CI
pnpm test # run the vitest suite once
pnpm test:watch # watch mode
pnpm test:coverage # run with a v8 coverage report (-> coverage/)
Tests live next to the code as *.test.ts and run on TypeScript source directly (no build step). The shared packages/* are covered by unit and integration tests — including a full Streamable-HTTP round-trip against a live server — and CI enforces a coverage floor on them. Each apps/* server ships an invariant suite that locks its tool surface (unique snake_case names, valid schemas and confirmation policies).
Every push and pull request to master runs CI: typecheck → build → tests with coverage. The coverage badge is regenerated from the run.
Authentication
You sign in once with your Microsoft account; the server then caches a refresh token and acquires access tokens silently from then on — no pasting, no 1-hour expiry. Sign-in uses your own Microsoft Entra ID app registration (free) so the servers act on your behalf.
1. Register an Entra ID app (one time)
-
Azure Portal → Microsoft Entra ID → App registrations → New registration. Name it anything; pick the Supported account types that fit (single-tenant, multi-tenant, and/or personal accounts).
-
Authentication → Add a platform → Mobile and desktop applications → add redirect URI
http://localhost, and set Allow public client flows to Yes (enables the--device-codefallback). -
API permissions → Add a permission → Microsoft Graph → Delegated permissions → add the scopes for the servers you use (then Grant admin consent if your tenant requires it):
Server Delegated scopes Calendar Calendars.ReadWriteContacts Contacts.ReadWriteOneDrive Files.ReadWrite.AllOutlook Mail.ReadWrite,Mail.SendSharePoint Sites.ReadWrite.AllAll servers also use
User.Read. (offline_accessis requested automatically for refresh.) -
Copy the Application (client) ID.
2. Sign in (one time per machine)
Set MICROSOFT_CLIENT_ID, then run the server's login command. A browser opens; after you consent, the token is cached under ~/.config/microsoft-mcp/:
export MICROSOFT_CLIENT_ID=<your-client-id>
npx -y ms-calendar-mcp login # opens the browser
npx -y ms-calendar-mcp login --device-code # headless: shows a code to enter
From then on the server refreshes tokens automatically. Use a non-default tenant with MICROSOFT_TENANT_ID (default common).
Advanced: supply your own token
To bypass the built-in flow, supply a pre-acquired Graph token directly:
- stdio: set
MICROSOFT_ACCESS_TOKEN(takes precedence over the cached sign-in). Good for quick tests — mint one withaz account get-access-token --resource https://graph.microsoft.com --query accessToken -o tsv. - HTTP: send
Authorization: Bearer <token>on eachPOST /mcprequest. Each request is stateless with its own token, so callers never share credentials — this is the model for hosted/remote deployments, which handle their own auth.
Running
stdio (e.g. Claude Desktop)
Each server is published to npm and runnable with npx — no clone or build. Sign in once first (npx -y ms-calendar-mcp login, see Authentication), then:
// claude_desktop_config.json
{
"mcpServers": {
"microsoft-calendar": {
"command": "npx",
"args": ["-y", "ms-calendar-mcp"],
"env": { "MICROSOFT_CLIENT_ID": "<your-client-id>" }
}
}
}
Or point at a local build instead of npm:
{
"command": "node",
"args": ["/abs/path/microsoft-mcp/apps/calendar/dist/index.js"],
"env": { "MICROSOFT_CLIENT_ID": "<your-client-id>" }
}
During development you can skip the build and run the TypeScript directly:
MICROSOFT_ACCESS_TOKEN=<token> pnpm --filter ms-calendar-mcp dev
Streamable HTTP
# build first, then:
PORT=3000 node apps/calendar/dist/index.js --http
# or, in dev:
pnpm --filter ms-calendar-mcp dev -- --http --port 3000
The server exposes POST /mcp (the MCP endpoint) and GET /healthz. Point any Streamable-HTTP MCP client at http://localhost:3000/mcp with an Authorization: Bearer header.
Transport selection
Resolved in this order: --stdio / --http flag → MCP_TRANSPORT=stdio|http → default stdio.
HTTP port: --port <n> → PORT → 3000.
Environment variables
| Variable | Used by | Description |
|---|---|---|
MICROSOFT_CLIENT_ID |
stdio | Entra ID app (client) ID for sign-in. Required for the login flow. |
MICROSOFT_TENANT_ID |
stdio | Tenant for sign-in: common (default), organizations, consumers, or a tenant ID. |
MICROSOFT_ACCESS_TOKEN |
stdio | Pre-acquired Graph token; overrides the cached sign-in when set. |
MICROSOFT_MCP_CACHE_DIR |
stdio | Override the token-cache directory (default ~/.config/microsoft-mcp). |
MCP_TRANSPORT |
both | stdio (default) or http. |
PORT |
http | Listen port (default 3000). |
MCP_HTTP_BODY_LIMIT |
http | Max request body size (default 50mb) for base64 uploads. |
MCP_DEBUG |
both | Any non-empty value enables debug logs (to stderr). |
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.