Use Our Email Tools From Your AI Assistant

Why a live check beats asking the AI raw (Why do I need your Email Deliverability MCP server when I can ask my questions directly to AI?)

Ask an AI about your domain without a tool connected, and one of three things happens - none of them is a DNS query:

1. It guesses. Your domain's DNS records and blacklist status are not in any training set - DNS zones are not crawled, and blacklists answer queries, they do not publish pages. A confident assessment of an SPF record the model has never seen is invention, not analysis.
2. It searches the web - and search finds pages about your domain, not your DNS. At best the AI paraphrases a third-party checker's cached page, secondhand and as old as that page's last render. Blacklists change hour to hour; DNS records change whenever someone edits them.
3. The careful ones decline - "I can't check live DNS." Accurate, and you still have no answer.

The advice layer is frozen too: a model trained before its cutoff may still recommend the pct= tag (removed from DMARC in RFC 9989) or quote sender rules that changed in 2024.

With this server connected, the guessing stops. That is what MCP email tools are for: instead of remembering, the AI measures - a DMARC MCP check reads the record that exists right now, a blacklist check queries all 24 lists at the moment you ask, a mail test analyzes a real message you just sent. Ground truth, in the chat, with a link to the full report.

claude.ai / Claude Desktop ✓ verified, with screenshots

These are the same product surface: connectors attach to your Claude account, and chats sync between the browser and the Desktop app - set it up once, use it in both. (Screenshots below are from the Desktop app on Linux; the browser dialogs are identical.)

1. Open claude.ai → your initials (bottom-left) → Settings → Connectors.
2. Click Add custom connector. Name it Postbox Tools, paste the URL above, and leave authentication on No sign-in (claude.ai auto-detects this - our server has open access by design).
3. Click Add. You will see all 8 tools listed with a permission setting each - Needs approval (Claude asks before each use) is a fine default; set the read-only checks to Always allow for smoother sessions.
4. In any new chat, open the tools menu near the message box, enable Postbox Tools, and ask away.

Claude Code (terminal) ✓ verified

One command, then the tools are available in your coding sessions (handy for checking a client's domain without leaving the terminal):

claude mcp add --transport http --scope user postbox-tools https://mcp.postboxservices.com/mcp

Confirm with claude mcp list - you should see postbox-tools ... ✓ Connected. Remove any time with claude mcp remove postbox-tools.

Cursor

Add the server to ~/.cursor/mcp.json (create the file if it does not exist), then restart Cursor. The tools appear in the Composer/Agent tool list:

{
  "mcpServers": {
    "postbox-tools": {
      "url": "https://mcp.postboxservices.com/mcp"
    }
  }
}

Standard Cursor remote-MCP config; verified run + screenshot coming.

ChatGPT ✓ verified, with screenshots

Use the web app (chatgpt.com) - in our testing (September 2026) it has the smoothest flow; the desktop app can manage MCP servers but lagged on exposing them in chats.

1. chatgpt.com → Plugins (sidebar) → Add ▾ → Add MCP server.
2. In the Create MCP App dialog: Name Postbox Tools, paste the server URL, and set Authentication: No authentication (it defaults to OAuth - our server needs none). Optionally add a description and our official icon: download the 128px PNG (5 KB). Tick the trust confirmation and Create.
3. First use shows a one-time Connect Postbox Tools consent - approve it.
4. Ask away. Tip: mentioning the tool by name ("...use the postbox_tools MCP") makes ChatGPT reach for it reliably; different models phrase the answer differently, but the underlying check is identical.

Grok ✓ verified, with screenshots

1. In Grok, open Settings → Connectors and choose Custom Connector.
2. Name: Postbox Tools, Server URL: the address above → Add Connector. Grok lists all 8 tools as enabled.
3. Ask in plain words - "Use Postbox Tools to check if mydomain.com is on any email blacklists" - and Grok calls the tool and answers from the live result.

The two-step full deliverability test works from Grok too: ask for a spam test, send one email from your own account to the disposable address it hands you, then ask for the score.

The 8 tools your AI gets

Every answer includes a link to the full report on the tool's own site - the AI gives you the summary, the report gives you every detail, shareable.

Data handling

We built this server the way we advise clients to build theirs, so here is the policy in plain words:

• Email files are never stored. analyze_eml processes the raw email in memory, returns the result, and discards it - nothing is written to disk, nothing is cached, and there is no way to retrieve a submitted file afterwards.
• Deliverability-test reports (create_mailtest) work exactly like the web tool: the report lives at its private link for 30 days, then is permanently deleted.
• Domain, IP and URL check results are cached briefly (up to 24 hours, keyed only by the input) so repeated lookups stay fast.
• Logs hold request metadata, not content: client IP, tool name, timestamps and fair-use counters, kept only as long as abuse control needs them. Message content never appears in any log.
• URL tracing is SSRF-hardened. Server-side fetching validates every redirect hop, not just the submitted URL - private, loopback and link-local ranges are blocked at each step, only http/https on standard ports is followed, and all fetching runs on isolated infrastructure.
• No accounts, no API keys, no stored conversations. We see the check input your AI sends - the domain, IP, URL or email file - and nothing else about your chat.

Frequently asked questions