Nel Veil

NEL VEIL MCP — free domain security checks for AI agents

Ask your agent *"can someone spoof email from example.com?"* and get a real answer in about ten seconds.

NEL VEIL MCP gives any MCP-compatible agent nine tools for checking a domain's public security posture: email spoofing (DMARC/SPF/DKIM), TLS weaknesses, HTTP security headers, publicly exposed files, subdomain-takeover risk, a combined scan, retrieval of an earlier scan, and compliance readiness against a register of 102 global frameworks.

Free. No API key. No signup.

Install

Claude Code — one command:

claude mcp add nel-veil -- npx -y nel-veil-mcp

Any MCP client — add this to your config file:

{

"mcpServers": { "nel-veil": { "command": "npx", "args": ["-y", "nel-veil-mcp"] } } }

Run it directly, to check it works:

npx -y nel-veil-mcp

It prints a ready line to stderr — something like nel-veil-mcp 0.2.1 ready — 9 tools, free passive tier. — and then waits for JSON-RPC on stdin. That is correct: it is a stdio server, not a CLI, and stdout carries the protocol, so nothing else is ever written there.

Requires Node 18 or newer.

30 seconds: is your domain spoofable?

After installing, ask your agent:

Check if example.com can be email-spoofed.

It calls check_email_spoofing. Real output for example.com at the time of writing:

Check if a domain can be email-spoofed — example.com

Score: 100/100 (done)

2 findings: [LOW] DMARC has no reporting address (rua/ruf), you have no visibility into spoofing [LOW] DKIM key may be 1024-bit (weak), 2048-bit recommended

Free check, built from public information. No port scanning and no exploit testing — though some checks (well-known path requests, admin-panel reachability) do more than passive reading; see nelprofessional.com/mcp. Active scanning requires verified domain ownership and runs only at nelprofessional.com. More: https://www.nelprofessional.com/mcp

That domain is in good shape. The finding people are most often surprised by is a DMARC policy of p=none — it looks configured, monitors everything, and blocks nothing.

Tools

Each returns a 0–100 score plus specific findings you can act on.

Compliance

check_compliance maps one passive scan onto a register of 102 frameworks across Canada, the United States, the EU/EEA, the United Kingdom, Asia-Pacific, Africa and the global standards — citing the specific provision each finding bears on.

Read the coverage field before repeating a score. Every framework reports how many of the scan categories it relies on were actually observed, as 9/16:

• A framework whose relevant checks did not run has a null score and the tier "Not assessed", with a reason. That is unknown, not clean.
• A clean result over fewer than half a framework's categories is "Partial", not "Strong".
• 26 of the 102 are "Not assessable" — an anti-money-laundering regime, a consumer-protection statute, a criminal offences provision or a securities disclosure rule imposes no technical safeguard an external scan could ever evidence. They are listed and explained, never scored.

These are indicative readiness signals from public-surface evidence. They are not an audit, not a certification and not legal advice, and every framework also carries a note on what it requires that no external scan can see — consent records, impact assessments, retention schedules, vendor agreements, board governance.

What this sends — stated plainly

Being precise here matters more than sounding safe, because this is the paragraph you rely on when deciding whether you may point a tool at someone else's domain.

No tool here does port scanning or exploit testing. Those are genuinely intrusive, they need the domain owner's permission, and they are deliberately not exposed over MCP at all.

But "passive" is not the same as "invisible", and several of these tools do more than read public records:

• `check_tls` connects to the domain, but only as a browser would. It makes one ordinary TLS handshake to port 443, from NEL's own infrastructure, and reads what the server presents: certificate issuer and subject, the validity window and how many days remain, the subject alternative names, and the negotiated protocol and cipher. Nothing else is contacted, and no third party is asked to test the target. It reports an expired certificate, one expiring within 14 or 30 days, a self-signed certificate, an untrusted chain, a certificate that does not cover the hostname, and a connection negotiated over TLS 1.0 or 1.1. What one handshake cannot tell you: it shows what the server chose for *that* connection, not everything the server would accept. It does not enumerate supported cipher suites, does not test for Heartbleed, POODLE, DROWN or RC4 support, and produces no letter grade. Those need the deep assessment described below.
• `check_subdomain_takeover` is not DNS-only. It resolves a small fixed list of common subdomain names and makes one HTTPS request to any that point at a known cloud host. It never claims or modifies anything.
• `check_exposed_files` requests a small fixed list of well-known paths (.env, .git/config and similar). It never brute-forces or fuzzes — the list does not grow or adapt — but these are paths a crawler would not request, so they are recognisable in a target's logs as a security check.
• `scan_domain` runs all of the above, and additionally checks whether common admin panels (/phpmyadmin/, /manager/html) are publicly reachable.

Everything above is information the domain publishes. None of it attacks anything. But it is more than ordinary crawling, so prefer running it against a domain you own or are authorised to assess.

Active scanning — port exposure, API probing, proof-of-concept checks, and the deep TLS assessment (tls_labs) that produces the Qualys SSL Labs grade along with the Heartbleed, POODLE, DROWN and RC4 tests — lives at nelprofessional.com, behind DNS-TXT proof that you control the domain. None of it is reachable from this MCP server. That is the right place for it, because a human proves ownership once and NEL can stand behind the result.

Rate limits. Ten single checks and three full scans per minute, per IP address. If you hit it, wait a minute.

Privacy. This package sends no API key, no email address, and no account identifier. Each request carries the domain you asked about and a nel-veil-mcp user-agent. Note that, like any HTTP service, NEL sees the IP the request came from — it is what the rate limit is keyed on.

Configuration

Development

These steps need the git repository, not the published npm package — the tarball ships only dist/, so src/ and scripts/ are not in it.

git clone https://github.com/NELPROINC/nel-veil-mcp.git

cd nel-veil-mcp npm install npm run build node scripts/mcp-smoke.mjs example.com

scripts/mcp-smoke.mjs launches the built server exactly as npx would, performs the MCP handshake, lists the tools and calls one for real — so it catches transport and schema problems a unit test would not.

About NEL VEIL

NEL VEIL is the security scanner behind NEL Professional, a marketplace connecting organisations with verified cybersecurity professionals. The same modules that power this MCP server power the scans on the site.

Full documentation: nelprofessional.com/mcp

License

MIT — see LICENSE.