Give Your AI Agent a Mailbox: Using InboxParse's Built-in MCP Server

AI agents can draft code, summarize documents, and answer complex questions — but most of them can't read your email. That's a problem, because email is still where the most important business context lives: customer requests, invoices, alerts, deal threads, and team updates.

InboxParse ships a production-ready Model Context Protocol (MCP) server. One URL, an OAuth login, and your AI agent can read, search, and act on real email threads — no custom integration code required.

What is MCP?

The Model Context Protocol is an open standard created by Anthropic for connecting AI models to external data sources and tools. Instead of manually copying data into a prompt, MCP lets your AI client call structured tools — like "search my inbox" or "get this email thread" — and receive typed responses it can reason over.

Think of it as a USB-C port for AI: one standard connector, many possible peripherals. InboxParse is one of those peripherals — it gives your agent eyes on your mailbox.

Setup Option 1: claude.ai (30 seconds)

If you use Claude on the web, this is the fastest path. No API keys, no npm, no config files.

1. Open Claude.ai → Settings → Integrations (or add it to a Project's MCP settings).
2. Add the server URL:

   https://inboxparse.com/api/mcp

3. Claude triggers an OAuth 2.1 flow — sign in to InboxParse, pick your workspace, and approve the scopes you want to grant.
4. Done. Claude can now call InboxParse tools directly in your conversations.

The OAuth flow uses PKCE, so no client secret is ever exposed. Tokens expire after 90 days and you can revoke them anytime from Console → MCP.

Setup Option 2: Claude Desktop & Cursor

For local AI clients, use the @inboxparse/mcp npm package. Add this to your claude_desktop_config.json (or Cursor's equivalent MCP settings):

{
  "mcpServers": {
    "inboxparse": {
      "command": "npx",
      "args": ["-y", "@inboxparse/mcp"],
      "env": {
        "INBOXPARSE_API_KEY": "ip_..."
      }
    }
  }
}

Replace ip_... with your API key from Console → API Keys. The package runs locally and proxies tool calls to the InboxParse API — nothing leaves your machine except authenticated API requests.

21 Tools at Your Disposal

Once connected, your AI client sees 21 tools organized into six categories. By default, only read scopes are granted — write scopes are opt-in.

Each tool returns structured JSON optimized for LLM consumption — emails come as clean Markdown with metadata, labels, AI summaries, and suggested responses already attached.

Real Example: Email Triage with Claude

Here's what it looks like in practice. You ask Claude to triage your inbox, and it chains tool calls together to give you an actionable summary:

User: Show me unread emails from the last 24 hours
      and flag anything urgent.

Claude calls: list_emails({
  date_from: "2026-03-21",
  format: "markdown",
  limit: 20
})

→ 14 emails returned

Claude calls: search_emails({
  query: "urgent OR overdue OR action required",
  mode: "hybrid"
})

→ 3 matches

Claude: Here's your morning triage:

🔴 Urgent (3)
  1. "Invoice #4821 overdue" from billing@acme.co
  2. "Action required: SSL cert expiring" from infra-alerts
  3. "Urgent: customer escalation" from support@bigcorp.com

📬 Everything else (11)
  4. Newsletter from Hacker News
  5. GitHub notification: PR #392 merged
  ...

Under the hood, the list_emails response looks like this:

{
  "data": [
    {
      "id": "msg_01jxxxxxxxxx",
      "subject": "Invoice #4821 overdue",
      "from": "billing@acme.co",
      "date": "2026-03-21T14:32:00Z",
      "labels": ["billing", "urgent"],
      "summary": "Payment of $2,400 for Q1 services is 15 days past due...",
      "body_markdown": "Hi team,\n\nThis is a reminder that invoice #4821..."
    }
  ],
  "pagination": { "total": 14, "page": 1, "per_page": 20 }
}

Notice the labels, summary, and body_markdown fields — InboxParse has already done the heavy lifting of parsing HTML email into clean Markdown and running AI classification. Your agent gets pre-processed, structured data instead of raw MIME.

Use Cases

Email triage and daily digests

Have your agent scan your inbox every morning and produce a prioritized summary. Urgent items surface first, newsletters get batched, and low-priority threads are silently archived. You start the day with a clear picture instead of a wall of unread messages.

Customer support automation

Connect your support mailbox and let your agent read incoming threads, apply labels based on topic and urgency, and draft responses using your knowledge base. The search_emails tool with mode: "semantic" is particularly powerful here — it finds related past conversations even when the exact wording differs.

Lead qualification

Point your agent at a sales inbox. It reads inbound emails, extracts company names, deal sizes, and intent signals, then routes qualified leads to your CRM. The combination of list_emails with date filters and search_emails with semantic search makes it straightforward to build a pipeline that catches opportunities you might otherwise miss.

Scoped Permissions

Security matters when giving an AI agent access to your email. InboxParse uses a fine-grained scope system:

• 7 read scopes are granted by default — your agent can read and search but not modify anything.
• 3 write scopes (mailbox:write, label:write, webhook:write) are opt-in — you explicitly approve them during the OAuth consent step.

You can review and revoke any MCP connection from your Console → MCP page. Each connection shows the client name, granted scopes, and last activity — so you always know what has access and when it was last used.

Coming Soon: Send Emails via MCP

Right now the MCP server is read-only by default, but we're actively building send support. Soon your AI agent will be able to compose and send emails, reply to threads, and forward messages — all through the same MCP connection. Write actions will require explicit opt-in via an email:send scope, so nothing changes for existing setups.

Stay tuned — this is landing in the next few weeks.

Get Started

InboxParse has a free plan with no credit card required. Connect a mailbox, add the MCP server URL to your AI client, and your agent can start reading email in under a minute.

• Create your free account
• Read the MCP server docs
• Full API reference