M365 Connector

v1.2.4 · Updated Mar 23, 2026

Install the Chrome extension, start the MCP server, and your AI agent gets full access to M365 — emails, files, Teams chats, calendar events, and people directory. No API keys, no app registrations. Tokens are captured passively from the browser.

MCP Tools

Five tools exposed to any MCP client:

Architecture

The MCP server runs as a long-lived process. The browser extension connects to it over WebSocket and handles all M365 API calls using tokens passively captured from the browser.

Install

1. Install the browser extension

Install from the Chrome Web Store:

Chrome Web Store → M365 Connector

Works on Chrome and Edge. After installing, the extension automatically captures auth tokens from your M365 session. Click the extension icon to verify — Substrate should show Valid.

2. Start the MCP server

Run with npx (no install needed):

npx m365connector

Or install globally:

npm install -g m365connector && m365connector

You should see:

[m365connector/ws] listening on ws://127.0.0.1:52365
[m365connector] MCP HTTP listening on http://127.0.0.1:52366/mcp

3. Configure your MCP client

Add to your MCP config (e.g. ~/.claude/mcp.json):

{
  "mcpServers": {
    "m365-connector": {
      "type": "http",
      "url": "http://127.0.0.1:52366/mcp"
    }
  }
}

Restart your MCP client to pick up the config.

Prerequisites

• Node.js 18+
• Chrome or Edge browser
• An M365 account (work or school)

Usage

Once connected, ask your AI agent to search M365 naturally:

> Search my email for messages about the quarterly review
> Find files related to Project Alpha
> Look up John in the company directory
> Search Teams chats about the deployment issue
> What meetings do I have tomorrow?

The agent uses search to find results, then open with a result's _readHandle to fetch full content. It can also call whoami, find_people, and count_people for directory and profile lookups.

Tools reference

search

Search across M365 data sources.

open

Read the content of a search result.

Returns content with a completeness field: full, partial, or snippet.

whoami

Get the current signed-in user's basic profile from Graph.

Returns display name, alias, email/UPN, title, department, office, phones, and best-effort manager and directReports info when directory scopes are available.

find_people

Find people via Graph directory query.

At least one filter parameter is required. Returns { filters, results, total }.

count_people

Count people matching directory filters without returning full results.

Returns { filters, count }.

Using as a Claude Code skill

M365 Connector ships with a Research skill that combines M365 search with web search (Lumina) to answer questions comprehensively. The skill gives Claude full knowledge of all 5 MCP tools plus a parallel dual-backend search strategy.

1. Install the extension + server — follow the install steps above.
2. Install the skill — use Claude Code's plugin system:

   /plugin marketplace add ttthree/m365connector-docs
   /plugin install research@m365connector-docs

   Or manually download SKILL.md:

   mkdir -p .claude/skills/research
   curl -fsSL https://raw.githubusercontent.com/ttthree/m365connector-docs/main/skills/research/SKILL.md \
     -o .claude/skills/research/SKILL.md

3. Use it — Claude Code discovers the skill automatically. Just ask:

   > What emails did I get from my manager today?
   > Summarize what's been happening on Project Alpha
   > Who is Alice Smith? What team is she on?
   > Find files related to the quarterly report

Environment variables

Security

• Both servers bind to 127.0.0.1 only — not accessible from the network
• WebSocket accepts connections only from chrome-extension:// origins
• Auth tokens are captured passively and stored in Chrome's local storage
• Read handles are server-issued opaque UUIDs with 30-minute TTL

Limitations

• Token refresh — requires an active M365 session in the browser (you must be logged in). The extension auto-refreshes tokens before expiry.
• Chat content — only search snippets are returned (full thread reading requires Graph Chat.Read scope).
• File content — returns extracted summaries, not full document bytes.

Troubleshooting