Graph MCP
An MCP to interact with Office 365 - Teams, mail, calendar.
Graph MCP
An MCP server that connects Claude to Microsoft Teams, Outlook Calendar, and Mail through the Microsoft Graph API.
What it does
Gives Claude access to 41 tools:
| Category | Tools |
|---|---|
| Auth | Check status, login (browser OAuth), logout |
| Chats | List chats, read/send messages, create chats, list members |
| Teams & Channels | List teams, list channels, read/send messages, list members, read/send replies |
| Calendar | List calendars, list/get events, create/update/delete events |
| List emails, read email, search emails, send email, reply, list/get attachments | |
| Meetings | Find meeting by Join URL, list/get transcripts, list recordings, get recording URL |
| Files | Browse OneDrive folders, search files, get content, upload files, create sharing links |
| Users | Search organization directory |
| Presence | Get/set your presence, get another user's presence |
| Search | Search messages across all chats and channels |
Prerequisites
You need an Azure App Registration:
-
Go to Azure Portal > App registrations > New registration
-
Set platform to Mobile and desktop applications
-
Set redirect URI to
http://localhost:3000/auth/callback -
Mark as Public client (no client secret needed)
-
Under API permissions, add these delegated permissions:
offline_access,openid,profile,User.ReadUser.ReadBasic.AllChat.Read,Chat.ReadWrite,ChatMessage.SendChannelMessage.Read.All,ChannelMessage.SendTeam.ReadBasic.All,Channel.ReadBasic.All,ChannelMember.Read.AllCalendars.ReadWriteMail.Read,Mail.SendPresence.Read,Presence.Read.All,Presence.ReadWriteOnlineMeetings.Read,OnlineMeetingTranscript.Read.All,OnlineMeetingRecording.Read.AllFiles.ReadWrite.All
Install
pip install graph-mcp
Or from a cloned repo:
pip install -e .
Setup
graph-mcp setup
This asks for your Azure Client ID and Tenant ID, then gives you the exact command:
claude mcp add graph -e AZURE_CLIENT_ID=your-id -e AZURE_TENANT_ID=your-tenant -- /path/to/graph-mcp
Paste it, start Claude Code, and ask Claude to log in. It opens your browser for OAuth sign-in — that's it.
Message and email formatting
The Teams message tools (graph_send_chat_message, graph_send_channel_message, and graph_reply_to_channel_message) and outbound mail tools (graph_send_mail, graph_reply_mail) now default to HTML mode and convert markdown to HTML automatically using Python-Markdown.
Examples:
**bold**→<strong>bold</strong>- itemlists →<ul><li>...</li></ul>`code`→<code>...</code>
If you already have raw HTML, pass it directly and it will be sent as-is.
For email, this means normal LLM-written markdown like **bold**, bullet lists, or inline code renders properly instead of showing raw markdown characters.
These tools also support optional top-level Graph API mentions, either in raw Graph format or a simplified shape such as:
[
{
"name": "Jane Smith",
"user_id": "ef1c916a-3135-4417-ba27-8eb7bd084193"
}
]
Use the corresponding <at id="0">Jane Smith</at> tag in the HTML body to trigger a real Teams mention.
How it works
Claude Code ──stdio──> graph-mcp ──HTTPS──> Microsoft Graph API
│
~/.graph-mcp/
tokens.enc (encrypted)
.key (auto-generated)
- Auth: OAuth2 Authorization Code flow with PKCE. Login opens your browser, a local callback server captures the token. No secrets stored in config.
- Token persistence: Tokens are encrypted with Fernet and stored in
~/.graph-mcp/tokens.enc. The encryption key is auto-generated on first use. Logins survive server restarts. - Token refresh: Access tokens are refreshed automatically before they expire. You only need to log in again if the refresh token itself expires.
- Rate limiting: Sliding window counter with exponential backoff. Respects
Retry-Afterheaders on 429 responses.
Configuration
All configuration is passed via environment variables (set in the MCP config's env block):
| Variable | Required | Default | Description |
|---|---|---|---|
AZURE_CLIENT_ID | Yes | — | From your Azure App Registration |
AZURE_TENANT_ID | No | common | Your Azure tenant ID, or common for multi-tenant |
GRAPH_REDIRECT_URI | No | http://localhost:3000/auth/callback | Must match Azure app registration |
GRAPH_TOKEN_ENCRYPTION_KEY | No | auto-generated | Fernet key for token encryption |
GRAPH_DEBUG | No | false | Enable verbose logging |
Troubleshooting
"Approval required" error during login Your Azure app is requesting scopes that aren't registered or need admin consent. Check the API permissions in the Azure portal match the list above.
403 Forbidden on specific tools The endpoint needs a permission that's not in your Azure app registration, or requires admin consent.
Login works but tools say "not authenticated"
Restart the MCP server (claude mcp restart graph) — it may be running an old version.
Disclaimer
This project is an independent open-source effort and is not affiliated with, endorsed by, or sponsored by Microsoft Corporation. Microsoft, Microsoft Teams, Outlook, Microsoft 365, Microsoft Graph, and Azure are trademarks of the Microsoft group of companies.
This software is provided "as is", without warranty of any kind. Use it at your own risk. The authors accept no liability for any damages, data loss, or security issues arising from the use of this software. You are responsible for complying with your organization's policies and Microsoft's API Terms of Use when using this tool.
This software accesses Microsoft services on your behalf using your own credentials and Azure app registration. Data retrieved from Microsoft Graph (emails, messages, calendar events, etc.) is passed to the LLM that invoked the tool. Be mindful of your organization's data handling policies when using this with cloud-hosted AI models.
License
MIT — see LICENSE.
相關伺服器
Kone.vc
贊助Monetize your AI agent with contextual product recommendations
Redmine MCP
Integrates Claude AI with the Redmine project management system to enhance project management tasks.
Pulsetic MCP Server
The Pulsetic MCP Server connects Pulsetic monitoring with AI agents and MCP-compatible tools, enabling direct access to uptime data, cron monitoring results, incident management workflows, and status page information through the Model Context Protocol (MCP). It allows teams to securely expose operational monitoring data in a structured format, making it easy to build AI-driven automation, monitoring assistants, and intelligent operational workflows without custom middleware.
Timergy MCP Server
Create scheduling polls and find the perfect meeting time. No auth required.
MCP Shrimp Task Manager
An intelligent task management system based on MCP, providing an efficient programming workflow framework for AI Agents with an optional web-based GUI.
JIRA
Interact with JIRA to search for issues using JQL and retrieve detailed issue information.
Apple Notes
Interact with Apple Notes using natural language on macOS.
Mcptix
A simple and powerful ticket tracking system with AI assistant integration.
Google Tasks
Interact with Google Tasks to manage your to-do lists and tasks.
Rootly
MCP server for the incident management platform Rootly.
Zapier
Connect your AI Agents to 8,000 apps instantly.