Agents are writing signup flows now. They're cleaning lead lists, wiring up CRMs, and building automations that used to take a sprint. And they all hit the same wall at some point: is this email address real?
An agent building a signup form needs to know whether [email protected] is a typo. An agent enriching a lead list needs to know which addresses will bounce before the campaign goes out. An agent guarding a free trial needs to spot the disposable domain someone is using to sign up for the fifth time.
Until now, the answer was "read our API docs and write the integration yourself." That works, but it's friction. So we removed it.
A remote MCP server
TinyValidator now runs a remote MCP server at:
https://tinyvalidator.com/mcp
It's Streamable HTTP with OAuth. There's nothing to install and no API key to copy around — your MCP client opens a browser window, you approve, and the tools show up.
Connecting takes one step in most clients:
- Claude Code:
claude mcp add --transport http tinyvalidator https://tinyvalidator.com/mcp - Claude.ai: Settings → Connectors → Add custom connector, paste the URL
- ChatGPT: Settings → Connectors (developer mode), add the URL
- Cursor: add
{ "url": "https://tinyvalidator.com/mcp" }undermcpServersin.cursor/mcp.json
The server exposes three tools:
get_pricing_for_volume— public, no authentication required. Your agent can quote what a job will cost before you even have an account. We did this on purpose: evaluation shouldn't require signup.verify_email— full verification of a single address: deliverability, disposable detection, catch-all status, role accounts, typo suggestions.verify_email_batch— up to 100 addresses in one call, for lists an agent is holding in memory.
The verification tools use the same account and credits as the REST API, authenticated through OAuth.
The engine behind the tools
The tools sit on the verification engine we recently rebuilt: RFC-aware DNS resolution, protocol-aware SMTP probes, and explicit catch-all detection, feeding an evidence-based score. The result payload treats suggestion, deliverability, disposable, catch_all, and role_account as distinct signals — so an agent can make a real decision instead of guessing from a single boolean. There's also an outcome-tracking endpoint so you can report back what actually happened to an email and help the scoring improve.
Not using MCP? Also fine
Plenty of agents integrate over plain HTTP, so the machine-readable surface covers that path too:
/llms.txt— a curated overview an agent can read in one passSKILL.md— when to use single vs. bulk verification, what each flag means, how to handle credentials/openapi.json— the full API contract- Markdown content negotiation on every docs page: send
Accept: text/markdown(or append?format=md) and you get structured text instead of HTML
An agent given nothing but the base URL can discover all of this and implement a working integration on its own.
n8n workflows
If your automation lives in n8n, we've built a community node: n8n-nodes-tinyvalidator, rolling out on npm now. Install it from Settings → Community Nodes, add your API key as a credential, and you get Verify (full verification) and Check (lightweight syntax/disposable/typo check) operations. It's also flagged as usable by n8n's AI agent nodes, so agents inside n8n can call it as a tool.
The one-paste setup
The /for-agents page has a copy-prompt button. It copies a short setup prompt — the MCP endpoint, the doc URLs, the REST fallback — that you paste into whatever agent you're using. The agent connects, verifies a test email, and confirms everything works. That's the whole integration.
Try it
The free tier is 1,000 validations, no credit card. Or skip the account entirely and ask your agent to call get_pricing_for_volume first — it'll tell you what your volume costs before you commit to anything.
Start at /for-agents. Your agent can take it from there.