25 security tools for AI agents: phishing links, exposure maps, DNS, SSL, PQC, history, pivots.
One address for any MCP client. No sign-in needed on this server.
9 read16 other
Grouped by tool name. At most: creates or changes data: records, files, settings.
Odel charges nothing per call. The publisher may charge for its own service.
Domain security reconnaissance for AI agents via the Model Context Protocol.
25 tools. No API key needed for non-commercial use. Phishing/smishing link checks, exposed admin page checks (admin screens, test servers, dev tools left open), DNS, SSL/TLS, post-quantum TLS readiness, HTTP security headers, email authentication, port scan, DNS propagation, reverse DNS, ASN/BGP, RDAP/WHOIS, subdomain discovery (CT logs), lookalike/typosquat detection, OWASP-mapped observable checks, brand-impersonation exposure, go-live readiness, domain change history — what changed since the last check — infrastructure history and pivots (past nameservers/IPs/issuers, and other domains seen on the same rare infrastructure), and a daily watch you can set from the agent.
Every result comes back interpreted, not just as raw JSON: a status, the key numbers, each issue with severity and confidence, and the concrete action to take. That is what an agent needs to tell a human what to do next.
Part of DechoNet. Every tool here is also a free web tool at dechonet.com — no sign-up — backed by error-fix guides. This package brings the same checks to AI agents.
No npx, no install. Point any MCP client that speaks Streamable HTTP at:
https://dechonet.com/mcp
Claude Desktop / Claude Code (claude mcp add --transport http dechonet https://dechonet.com/mcp), Cursor, and other HTTP-capable clients connect directly. The remote endpoint also serves 3 curated prompts (audit_domain, monitor_setup, investigate_changes) and 2 reference resources.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"dechonet": {
"command": "npx",
"args": ["-y", "dechonet-mcp"]
}
}
}
Config file location:
~/Library/Application Support/Claude/claude_desktop_config.json%APPDATA%\Claude\claude_desktop_config.jsonRestart Claude Desktop. You'll see the DechoNet tools icon in the input area.
# Via npx (no install needed)
npx dechonet-mcp
# Or install globally
npm install -g dechonet-mcp
| Tool | Description |
|---|---|
phishing_link_check | Is this link a scam? — paste a text message or URL; follows shorteners, checks domain age, Korean brand impersonation, APK downloads, phishing feeds |
exposure_map | Admin pages open to the internet? — subdomains from CT, each opened once: admin screens, staging, dev tools, VPN logins, default pages, directory listings (counts; per-host detail for the verified owner on the web report) |
security_scan | Comprehensive scan — 9 checks in parallel, 0-100 Health Score, A-F grade |
dns_lookup | DNS records + DNSSEC + SPF/DMARC validation |
ssl_check | SSL/TLS certificate, chain, expiry, HSTS, CT history, A-F grade |
http_security | HTTP redirect trace + 10 security headers audit, A-F grade |
email_auth | SPF, DMARC, DKIM, BIMI, MTA-STS, DANE + blacklist check |
port_scan | Open TCP ports with service identification |
dns_propagation | DNS propagation across 8+ global resolvers |
reverse_dns | PTR record + FCrDNS verification |
asn_lookup | ASN/BGP network identification + abuse contact |
whois_lookup | RDAP/WHOIS domain registration data |
subdomain_discovery | Passive subdomain enumeration from CT logs, operational-name flags (dev/staging/admin), wildcard detection |
lookalike_domains | Typosquat variants that are actually registered — homoglyph, affix (brand-login), TLD swap, keyboard slips — with the domain's own defensive registrations separated out |
owasp_check | OWASP-mapped checks that can be observed passively (headers, TLS, exposed files), honest about what is out of scope |
impersonation_exposure | Brand impersonation exposure grade: third-party lookalikes + exposed operational subdomains + wildcard certs |
golive_check | Go-live readiness — DNS, propagation, SSL, HTTP, registration in one READY / CAUTION / NOT READY verdict |
domain_history | Past nameservers, A/AAAA, MX, certificate issuer, registrar (and an IP's ASN/PTR) with first and last seen — from DechoNet's own observations, not passive DNS |
infrastructure_pivot | Other domains DechoNet saw with the same nameserver, IP, MX, issuer or registrar. Needs a free API key (DECHONET_API_KEY); shared infrastructure returns a count only |
domain_changes | What changed since the last check — status, grade, issuer, DNS, issues. Works from lookup history even without a watch |
pqc_readiness | Post-quantum TLS readiness: hybrid X25519MLKEM768 key exchange, whether the origin server or a CDN edge provides it, certificate signature algorithms. External indicator, not a full PQC audit |
watch_domain | Register a domain for daily re-checks (SSL, DNS, HTTP, RDAP) so domain_changes accumulates a timeline. The only tool that writes; idempotent |
ip_info | Public IP, ISP, ASN, proxy detection |
email_header_analysis | Email delivery route tracing + auth results |
subnet_calc | CIDR subnet calculator (offline) |
Every tool response ends with a link to the full interactive report on dechonet.com for the human behind the agent.
Once connected, try asking Claude:
For a self-hosted HTTP/SSE bridge (the hosted remote endpoint above is usually simpler):
npm run start:sse
# Server runs on http://localhost:3100
# SSE endpoint: http://localhost:3100/sse
npm install
npm run build # TypeScript → build/
npm run dev # Run with tsx (stdio)
npm run start:sse # Run SSE server
The MCP server calls DechoNet's public API (https://dechonet.com/api/util/*) — the same backend as the dechonet.com web tools — and returns structured results with:
All data comes from public sources (DNS, HTTP headers, SSL certificates, CT logs, RDAP). Passive by design: no active exploitation, and a registered lookalike is reported as a fact to verify, never as an accusation.
| Variable | Default | Description |
|---|---|---|
DECHONET_URL | https://dechonet.com | API base URL |
DECHONET_LOCALE | en | Response language (en or ko) |
DECHONET_API_KEY | — | Optional free API key from your DechoNet account: your own rate-limit bucket, and required for infrastructure_pivot |
PORT | 3100 | SSE server port |
Free use (with or without a key) is for non-commercial purposes from November 8, 2026 — personal learning and research, checking your own domains, non-profit work. Using DechoNet inside a paid product, for clients, or regularly for company work needs MCP·API Pro (dechonet.com/pro); until payments open, ask at support@dechonet.com. Full terms: dechonet.com/terms §4.
DechoNet is a domain security reconnaissance platform — distributed observation, cross-verification. The same diagnostics ship three ways: free web tools, error-fix guides, and this MCP server.
MIT licensed. Issues and PRs welcome.