Postgate AI - Microsoft Exchange MCP Server

Máy chủ MCP đa người dùng được lưu trữ dành cho Microsoft Exchange - bao gồm Exchange Server tại chỗ và hybrid (EWS): hơn 70 công cụ cho thư, lịch, danh bạ và tác vụ qua SSE/Streamable HTTP với OAuth 2.0/PKCE.

Tài liệu

MCP server for Microsoft Exchange Server SE: on-prem (EWS) & hybrid.

Postgate AI exposes Exchange mailboxes as an MCP server, with 90+ tools over Streamable HTTP, OAuth 2.0/PKCE, and delegated per-user permissions. Mailboxes in your own network connect via EWS, which covers Exchange Server Subscription Edition (SE) plus Exchange 2019 and 2016. Mailboxes in Exchange Online are reached via Microsoft Graph. You can run it fully managed in Germany or self-hosted under license.

Architecture

Seven things to know about the technology

The MCP standard

Model Context Protocol is the open standard that AI assistants use to reach external systems. Postgate AI exposes Microsoft Exchange as an MCP server for Claude, ChatGPT, Perplexity, Mistral Vibe, and any other MCP-capable client. We fully support the July 28, 2026 revision. Older clients keep working just as before, so there is nothing for you to change.

Transport: Streamable HTTP (July 28, 2026 revision)

We implement the July 28, 2026 MCP revision on the server side, with a single endpoint: /mcp. The new revision drops three things that all used to cost time: the session ID, the handshake before the first tool call, and the stream that each client had to keep open. Every request now stands on its own. For self-hosted setups, that means you can run several nodes behind a plain load balancer without session affinity and add another node as soon as load grows. A restart no longer cuts anyone off. Every route uses TLS, and no client needs Server-Sent Events anymore.

Sign-in: OAuth 2.0/PKCE, Entra ID or AD FS

Clients prove their identity via OAuth 2.0 with PKCE instead of relying on trust-based headers. Users onboard themselves with a Microsoft sign-in or a one-time code from the portal. Out of the box, sign-in runs through Microsoft Entra ID, which also fits a hybrid setup. If you don’t want any cloud sign-in, point us to your own AD FS as the issuer, and we validate tokens from your own environment. Mailbox access uses native Windows authentication (NTLM). Once Exchange trusts your own issuer, it can use modern auth with a per-user token instead, and no domain password is stored in the vault.

Per-user vault

Exchange credentials are stored per user in a vault, encrypted with RSA-4096. The vault sits in Germany when we run the service for you, or in your own network when you self-host. Access is delegated and uses existing Exchange permissions, with no service account and no application impersonation.

Connected via EWS and Microsoft Graph

Mailboxes on an Exchange Server in your own network connect via EWS. This works with Exchange Server SE, 2019, and 2016. Mailboxes in Exchange Online are reached via Microsoft Graph. Both run side by side in the same setup, and each organization decides which one it uses. You don’t need a cloud mailbox.

Several nodes behind one load balancer

The current MCP revision works without session affinity. Several instances can therefore run side by side, and each request goes wherever there is capacity. When the number of users grows, you add another node.

90+ tools

The tools cover mail, calendar, contacts, tasks, folders, attachments (including PDF/DOCX/XLSX extraction), threads, the global address list, free/busy times, and semantic search. Your entire mailbox becomes a toolbox.

Integration

One MCP server, three ways to use it

AI assistants

Use Claude, ChatGPT, Perplexity, and Mistral Vibe on desktop and mobile. Gemini works via Gemini Enterprise or Google Antigravity.

Agent platforms

Microsoft Copilot Studio plugs Postgate AI in as an MCP tool. Your Copilot agent then works with on-premises mailboxes. The same goes for n8n and for any other platform that speaks MCP with OAuth.

Custom agents

Any MCP-capable runtime can use the tools, from a Python script to a full agent platform.

Step-by-step guides: Claude, ChatGPT, Gemini, Perplexity, Mistral Vibe →

FAQ

Frequently asked questions

What is an MCP server for Microsoft Exchange?

It’s a server that makes Exchange mailboxes accessible to AI assistants via the Model Context Protocol (MCP). Emails, calendar, contacts, and tasks become standardized tools that Claude, ChatGPT, Mistral Vibe, or custom agents can call, using each user’s own permissions.

Does the MCP server work with Exchange on-premises?

Yes, that’s the core of it. Postgate AI connects via EWS to Microsoft Exchange Server SE, 2019, and 2016, including hybrid environments. You can run it fully managed or self-hosted under license, and in both cases nothing is installed on clients.

Can I use the MCP server in Copilot Studio or n8n?

Yes, both work. In Microsoft Copilot Studio you add Postgate AI as an MCP tool (Streamable HTTP, OAuth 2.0 with dynamic discovery) and your Copilot agent then reaches on-premises mailboxes. n8n works the same way. In both cases every user works with their own Exchange permissions.

How do users authenticate?

Clients authenticate via OAuth 2.0 with PKCE. Each user’s Exchange credentials are stored encrypted in the per-user vault. Every access is delegated and uses that user’s existing Exchange permissions.

Exchange as an MCP server, ready to try in minutes.

Try the fully managed plan free for 14 days: connect your EWS URL, pair a client, and start using the tools. Contact us for a self-hosted license.