Mailverdict MCP Server

mailverdict

Email intelligence API: disposable/burner detection, role-account detection, free-provider classification, MX validation, and typo suggestions. Keyless and free at the entry tier, built to be the email-check dependency that coding agents reach for when they scaffold a signup flow.

API: https://api.mailverdict.dev · npm: mailverdict · Freshness feed ("Burner Feed"): GET /v1/changes

Why this exists

Every SaaS signup form needs the same three answers — is this a burner? is this a person or a mailbox like info@? did they typo their domain? — and the current options are stale GitHub lists you have to vendor yourself or paid validators gated behind signup. The value here is not the code (it's small on purpose): it's the maintained dataset — burner domains churn daily, and this repo rebuilds its dataset from every maintained public source on a daily schedule, plus its own discovery over time.

Quickstart

npm install
npm run build:data   # fetch + merge upstream lists into data/dataset.json
npm test
npm run dev          # http://localhost:8787

Requires Node >= 20.

curl "localhost:8787/v1/check?email=jane@mailinator.com"

{
  "email": "jane@mailinator.com",
  "user": "jane",
  "domain": "mailinator.com",
  "tag": null,
  "normalized_email": "jane@mailinator.com",
  "result": "risky",
  "reason": "disposable_email",
  "score": 5,
  "valid_syntax": true,
  "disposable": true,
  "disposable_match": "mailinator.com",
  "role": false,
  "free": false,
  "mx_found": true,
  "mx_records": ["mail2.mailinator.com"],
  "did_you_mean": null,
  "accept_all": null
}

Response follows the Kickbox/Emailable industry convention (result, reason, bare booleans, did_you_mean = full corrected address, score 0–100) so code written against those APIs ports with near-zero edits.

Endpoints: GET /v1/check?email= · GET /v1/check/{email} · POST /v1/check (batch ≤100) · GET /v1/domain/{domain} · GET /v1/changes?since=24h (freshness feed) · GET /v1/meta · /llms.txt · /openapi.yaml. Add mx=false to skip the DNS lookup.

Privacy-conscious default: if you only need burner/typo screening, check the domain (/v1/domain/{domain}) — no email address leaves your system. Use the full email check when you need role detection or normalization.

SDK: living data with a static safety net

sdk/ packages the integration story: call the live API, fall back to a bundled offline snapshot on timeout/failure — fresher than a vendored static list, but never a hard runtime dependency in your signup path.

import { createClient } from 'mailverdict'
const check = await createClient().checkEmail(input)
if (check.disposable) reject()          // check.source: 'live' | 'snapshot'

Published on npm as mailverdict; see sdk/README.md.

MCP server

Hosted, keyless remote server — point any MCP client at:

POST https://api.mailverdict.dev/mcp

Stateless Streamable HTTP (JSON-RPC 2.0), no signup or key. Tools: check_email, check_domain.

A stdio server also ships for local use:

npm run mcp   # stdio transport; tools: check_email, check_domain

Register the local one in Claude Code: claude mcp add mailverdict -- npx tsx src/mcp.ts (run from this directory).

Architecture

pipeline/build.ts     daily: fetch upstream lists -> merge/dedupe -> apply
                      allowlist -> data/dataset.json + data/meta.json
src/lib/*             pure logic: parse, classify, suggest, MX-over-DoH
src/app.ts            Hono app (runtime-agnostic)
src/server.ts         Node entry (self-hosting)
src/worker.ts         Cloudflare Workers entry (production: api.mailverdict.dev)
src/mcp.ts            MCP stdio server for coding agents
.github/workflows/    daily dataset refresh + auto-commit

DNS is done over DoH (cloudflare-dns.com) so the same code runs on Node and edge runtimes. Subdomains match their parents (foo.mailinator.com → mailinator.com).

Data sources & attribution

The dataset aggregates these excellent permissively-licensed community lists — see data/meta.json for per-source counts, status, and license on every build (unlicensed and GPL lists are deliberately excluded):

• disposable-email-domains (CC0; blocklist + pinned allowlist — upstream deleted allowlist.conf 2026-04)
• disposable/disposable (MIT; large daily-CI aggregator)
• groundcat/disposable-email-domain-list (MIT; MX-validated + maintained allowlist)
• FGRibreau/mailchecker (MIT)
• wesbos/burner-email-providers (MIT)
• 7c/fakefilter (BSD-3-Clause; bot-updated)
• unkn0w/disposable-email-domain-list (MIT; hand-curated)
• Kikobeats/free-email-domains (MIT; HubSpot-derived)
• mixmaxhq/role-based-email-addresses (MIT; role local-parts), plus curated extras in pipeline/roles.ts

Merge precedence (important — the free-provider list contains temp-mail domains and typo-squats like mailinator.com and gmial.com): allowlists rescue from disposable, then disposable wins over free.

Roadmap

mailverdict is built agent-first: the first call works with zero signup, the docs are operable by coding agents (/llms.txt, OpenAPI), and the dataset's freshness is provable via the Burner Feed rather than claimed.

• Keyless free API — the first call works with zero signup
• Industry-compatible response shape (Kickbox/Emailable convention: result, reason, did_you_mean, bare booleans)
• llms.txt (agent-operable: endpoint-first, curl-able), OpenAPI spec
• MCP server — local stdio and hosted keyless remote (Streamable HTTP) at /mcp
• Daily automated dataset refresh + auto-deploy
• Freshness changelog + /v1/changes feed (clean-build diffs only; 90-day retention)
• SDK with bundled snapshot fallback, published on npm (mailverdict)
• Production deployment on Cloudflare Workers (api.mailverdict.dev)
• Official MCP registry listing (dev.mailverdict/mailverdict)
• Per-IP rate limiting
• Detection-lag benchmark: measure and publish how quickly new burner domains enter the dataset
• Better Auth plugin (keyless check in the before-create hook)
• Framework snippets (Zod .refine(), React Hook Form, server actions)
• Optional API keys for high-volume / commercial use

Status

v0.1.1 — live at https://api.mailverdict.dev (Cloudflare Workers), with a hosted keyless remote MCP server at /mcp and an official MCP registry listing. The dataset refreshes and redeploys daily via GitHub Actions, and the mailverdict SDK is published on npm. 41 tests cover the core logic, HTTP layer, MCP handler, and SDK live/fallback paths.

Disclaimers, privacy & fair use

• Advisory signals, not verdicts. Classifications aggregate public blocklists and heuristics; false positives happen (domains churn, upstream lists disagree). Don't hard-block users on a single boolean — combine with your own signals and give flagged users a path forward.
• Wrongly listed? If your domain is misclassified, open a misclassified-domain report. Allowlist corrections ship with the next daily build, and fixes are submitted upstream when a source list is at fault.
• Privacy. No accounts, no logging of email addresses, nothing stored per request. MX checks send only the domain (never the local part) to a public DNS-over-HTTPS resolver. Traffic metrics are aggregate and PII-free.
• No SLA. The hosted API is free and best-effort while in beta. If your signup flow must never block on us, use the SDK (it falls back to a bundled offline snapshot) or self-host — see docs/deploy.md.
• Fair use. Built for signup and form validation inside applications. Do not use it to bulk-clean purchased or scraped mailing lists; per-IP rate limits apply and abusive traffic gets blocked.

Commercial use & higher limits

The hosted API is free with per-IP rate limits. Need higher throughput, an SLA, or commercial terms? Email hello@mailverdict.dev.

License

MIT — see LICENSE; it covers the code and dataset pipeline. The hosted service at api.mailverdict.dev is provided "as is", without warranty of any kind. The aggregated dataset sources carry their own permissive licenses, recorded per-source in data/meta.json.