domain-security-mcp-server

# domain-security-mcp-server

> An [MCP](https://modelcontextprotocol.io) server that lets an AI agent audit the **email and domain security** of any domain — SPF, DKIM, DMARC, MTA-STS, TLS-RPT, BIMI, DNSSEC, DNS, TLS/SSL and WHOIS — in plain language. **No API keys required.**

[![ci](https://github.com/OrtaMarco/domain-security-mcp-server/actions/workflows/ci.yml/badge.svg)](https://github.com/OrtaMarco/domain-security-mcp-server/actions/workflows/ci.yml)
[![npm](https://img.shields.io/npm/v/domain-security-mcp-server)](https://www.npmjs.com/package/domain-security-mcp-server)
[![MCP](https://img.shields.io/badge/MCP-server-blue)](https://modelcontextprotocol.io)
[![TypeScript](https://img.shields.io/badge/TypeScript-strict-3178c6)](https://www.typescriptlang.org/)
[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](./LICENSE)

Built on the **v2 MCP SDK**: the server speaks the **2026-07-28** protocol
revision and keeps accepting 2025-era clients (Claude Desktop, Claude Code,
Cursor) from the same factory — the era is negotiated per connection, so there
is nothing to configure on either side.

Ask Claude *"Is acme.com protected against email spoofing?"* and it runs a full
authentication audit and hands you a graded report with prioritised fixes —
instead of you pasting a domain into five different web tools.

```
> Is ortamarco.me protected against email spoofing?

  email_auth_audit(domain="ortamarco.me")

  Grade: A (95/100) · MX: present
  ✅ SPF ends in '-all' (hard fail). 3/10 DNS lookups.
  ✅ DMARC policy is enforced ('p=reject').
  ✅ DKIM key found for selector: google.
  Top recommendation: add a TLS-RPT record for delivery-failure reports.
```

---

## Why this exists

The email-security ecosystem is full of single-purpose web checkers (SPF here,
DMARC there, WHOIS somewhere else) and the few MCP equivalents are locked behind
paid API tokens. This server brings the whole **deliverability & domain-security
toolkit** to any MCP client, key-free, with one headline workflow tool that does
the synthesis for you.

It is the agent-facing companion to the network tools at
[ortamarco.me](https://ortamarco.me) and shares the same battle-tested core
(public-resolver DNS, host validation, timeouts).

## Tools

| Tool | What it does |
|---|---|
| **`email_auth_audit`** ⭐ | One-call SPF + DKIM + DMARC + MX audit → 0–100 score, A–F grade, prioritised fixes |
| `spf_check` | Parse SPF; recursively count DNS lookups vs the RFC 7208 limit of 10; flag `+all`/`?all` |
| `dmarc_check` | Parse DMARC policy (`p`, `sp`, `rua`, `pct`, `aspf`/`adkim`) with warnings |
| `dkim_check` | Probe `<selector>._domainkey` keys (supply selectors or use common ones) |
| `mta_sts_check` | Validate the `_mta-sts` TXT **and** the `.well-known/mta-sts.txt` policy + mode |
| `tls_rpt_check` | Check the `_smtp._tls` TLS-RPT record |
| `bimi_check` | Check the `default._bimi` BIMI record |
| `dnssec_check` | DS/DNSKEY presence + DNSSEC `AD` validation flag (via DoH) |
| `dns_lookup` | All record types (A/AAAA/CNAME/MX/NS/TXT/SOA) via public resolvers |
| `ssl_certificate` | TLS cert issuer, validity window, days-to-expiry, SANs, fingerprint |
| `whois_lookup` | Registrar, dates, name servers, status (raw port-43 WHOIS, IANA-resolved) |
| `reverse_dns` | PTR records for an IP |
| `ip_geolocation` | Offline IP geolocation ([DB-IP Lite](https://db-ip.com)) + reverse DNS |
| `mx_lookup` | Mail servers (MX) with priority and resolved IPs |
| `caa_check` | Which CAs may issue TLS certificates (CAA records) |
| `blacklist_check` | IP/domain against open-access email DNSBLs |
| `dns_propagation` | Compare a record across 5 public resolvers worldwide |
| `http_security_headers` | Grade a site's HSTS, CSP, X-Content-Type-Options, X-Frame-Options, Referrer-Policy, Permissions-Policy and COOP |
| `analyze_email_headers` | Parse raw headers → SPF/DKIM/DMARC verdicts + Received hop chain with delays |

Every tool is **read-only**, declares an `outputSchema` and returns
`structuredContent` (validated by the SDK) alongside human-readable Markdown
(default) or JSON (`response_format="json"`), plus actionable error messages.

## Install

Requires **Node.js 20.18+**. Nothing to clone — every MCP client can run it with `npx`.

## Use it with Claude Code

```bash
claude mcp add domain-security -- npx -y domain-security-mcp-server
```

## Use it with Claude Desktop or Cursor

Add to `claude_desktop_config.json` (or `~/.cursor/mcp.json`) — see [`examples/`](./examples/claude_desktop_config.json):

```json
{
  "mcpServers": {
    "domain-security": {
      "command": "npx",
      "args": ["-y", "domain-security-mcp-server"]
    }
  }
}
```

On Windows use `"command": "cmd"` with `"args": ["/c", "npx", "-y", "domain-security-mcp-server"]`.
Restart the client, then ask: *"Audit the email security of stripe.com."*

## Self-host (HTTP transport)

The same server speaks stateless **Streamable HTTP** for remote or multi-client
use. One endpoint serves both protocol eras and there is no session state, so no
`Mcp-Session-Id` header is issued or expected.

```bash
TRANSPORT=http npx -y domain-security-mcp-server
# POST JSON-RPC to http://127.0.0.1:3000/mcp   ·   health at /healthz
```

It is **safe by default**: it binds to `127.0.0.1` and only accepts `localhost`
`Host` and `Origin` headers, which blocks DNS-rebinding attacks from a web page.
To expose it — for example behind Coolify or Traefik — opt in explicitly:

| Variable | Default | Purpose |
|---|---|---|
| `TRANSPORT` | `stdio` | `http` to serve Streamable HTTP |
| `PORT` | `3000` | Listening port |
| `HOST` | `127.0.0.1` | Bind address; `0.0.0.0` to accept remote connections |
| `ALLOWED_HOSTS` | — | Comma-separated hostnames the `Host` header may carry (e.g. `mcp.example.com`) |
| `ALLOWED_ORIGINS` | — | Comma-separated origins allowed to call from a browser |
| `MCP_AUTH_TOKEN` | — | If set, every request needs `Authorization: Bearer <token>` |

Binding to a non-loopback address without `ALLOWED_HOSTS` or `MCP_AUTH_TOKEN`
works, but the server says so on stderr. With Docker (the image sets `HOST=0.0.0.0`):

```bash
docker build -t domain-security-mcp .
docker run -p 3000:3000 -e ALLOWED_HOSTS=mcp.example.com -e MCP_AUTH_TOKEN=change-me domain-security-mcp
```

## Security

The tools reach out to hosts that the *caller* names, so every outbound
connection is screened against server-side request forgery:

- Private, loopback, link-local (cloud metadata), shared, multicast and reserved
  addresses are refused in every spelling, including IPv4 embedded in IPv6
  (`[::ffff:169.254.169.254]`).
- The check happens **at connect time**, on the address the socket is actually
  about to use, so DNS rebinding and names only an internal resolver knows are
  refused too. Redirects are followed by hand and every hop is re-checked.
- Response bodies, redirects, WHOIS referrals and every network call are capped
  and time-limited.

Found a problem? Please open a [private security advisory](https://github.com/OrtaMarco/domain-security-mcp-server/security/advisories/new).

## Develop

```bash
npm run dev      # tsx watch (stdio)
npm run inspect  # open the MCP Inspector against the built server
npm run build     # type-check + emit dist/
npm run typecheck # type-check only
npm test          # offline unit tests: SSRF guard, SPF/DMARC/DKIM scoring,
                  # header parsing, HTTP transport defaults
npm run smoke     # call all 19 tools on BOTH protocol eras and validate
                  # structuredContent against each tool's outputSchema
```

[`evals/`](./evals/) holds a 10-question LLM evaluation set (stable, verifiable)
and instructions for running it — see [`evals/README.md`](./evals/README.md).

## How it works

```
src/
├── index.ts        # transport selection (stdio | http), v2 SDK entry points
├── server.ts       # factory: registers every tool on one McpServer
├── core/           # pure logic, no MCP coupling — reusable & testable
│   ├── validate.ts # input validation and the address classifier
│   ├── netguard.ts # connect-time SSRF guard, redirect-safe fetch, capped bodies
│   ├── dns.ts      # public-resolver DNS + DoH client
│   ├── net.ts      # MX, CAA, DNSBL and propagation checks
│   ├── tls.ts      # certificate inspection and trust
│   ├── whois.ts    # port-43 WHOIS with IANA/registrar referral
│   ├── http.ts     # security-header grading
│   ├── geoip.ts    # offline IP geolocation on DB-IP Lite (each file read on first use)
│   ├── email-headers.ts  # raw header parsing and hop timing
│   └── email-auth.ts  # SPF/DKIM/DMARC/MTA-STS/TLS-RPT/BIMI/DNSSEC + scoring
└── tools/          # thin MCP wrappers (Zod schemas, descriptions, formatting)
```

The `core/` layer is deliberately free of any MCP types, so the exact same logic
powers both this server and the web tools on ortamarco.me.

## Credits

<a href='https://db-ip.com'>IP Geolocation by DB-IP</a>. `ip_geolocation` uses the
free DB-IP "IP to City Lite" database, licensed under
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) and installed as the
[`@ip-location-db/dbip-city-mmdb`](https://www.npmjs.com/package/@ip-location-db/dbip-city-mmdb)
package; the credit also appears in the tool's description and Markdown output. The
database has no time zone, so `time_zone` is estimated from the coordinates with
[`@photostructure/tz-lookup`](https://github.com/photostructure/tz-lookup) (CC0).

## License

MIT © [Marco Orta](https://ortamarco.me)