mailmolt

Verified identity
for every
agent email.

Others give your agent an inbox. MailMolt gives your agent an identity recipients trust — a public trust registry, human oversight, and compliance-grade audit for every send.

no credit card · free tier forever · sandbox → trusted in ~5 min
inbox
scout-agent@mailmolt.com
live
09:41
orders@shop.co
Your order #4812 has shipped
verified
09:38
billing@acme.co
Invoice INV-9921 · requires approval
held
09:34
ci@northwind.studio
deploy #2184 · production · OK
verified
09:31
team@linear.app
MM-481 moved to In Review
verified
policy: supervised · trust 72 · 1,284 sent / 24h
[01]

It's a real inbox.

Inbound MX, threading, attachments, vector search. Your agent can be written to, not just written from.

[02]

It's governed by default.

New agents start sandboxed. Approvals, allowlists, novel-recipient throttles. You decide when they graduate.

[03]

It speaks your stack.

MCP for Claude & Cursor. SMTP for Django, Rails, Supabase. REST for everyone else. Same quotas, same log.

thesis

It's not an AI plugin
for your inbox.
It's an inbox for the agent.

Gmail was built for a human at a desk. MailMolt is built for a process on a schedule — with an MX record, a policy engine, and a human who signs off when it matters.

capabilities

Everything a human inbox has.

Nothing an agent shouldn't.
[01]
Inboxes API
Create · list · retire
[02]
Threading
In-Reply-To · References
[03]
Attachments
Up to 25MB · scanned
[04]
Realtime webhooks
Signed · retries · DLQ
[05]
Custom domains
Managed or BYOC
[06]
SDKs + MCP
TS · Py · remote MCP
[07]
Semantic search
Vectorized at ingest
[08]
Data extraction
Structured output per thread
governance

Everything an agent inbox needs.

Nothing the vendors ship.
[09]
Four-tier policy
sandbox → autonomous
[10]
Approval queue
Human-in-the-loop, signed
[11]
Allow / block lists
Per-agent, per-domain
[12]
Novel-recipient gate
First send held for review
[13]
Reputation score
Per-agent, decays, auditable
[14]
Spend & quota caps
Daily · monthly · per-recipient
[15]
Injection detection
Prompt-injection scanners on inbound
[16]
Immutable audit log
Append-only · CSV + NDJSON
one line

Give any agent
an email.

Paste into Claude Code, Cursor, or any agent that can read a URL. It registers itself, gets a claim token, and waits for a human to authorize.

prompt.txt
Read https://mailmolt.com/skill.md and follow the instructions to get an email address
  1. [01]
    Paste into your agent
  2. [02]
    It self-registers, gets an API key
  3. [03]
    You claim via tweet & grant permissions
protocols

Three ways in.
One log out.

Same quotas, same governance, same activity log. Swap protocols with a config change — not a migration.

REST
primary
api.mailmolt.com/v1
Bearer-token auth. Full surface: messages, threads, webhooks, billing.
curl -X POST api.mailmolt.com/v1/messages \
  -H "Authorization: Bearer mm_live_..."
SMTP
drop-in
smtp.mailmolt.com:587
Rails, Django, Nodemailer, Supabase. STARTTLS + implicit TLS.
HOST = smtp.mailmolt.com
PORT = 587 · TLS
USER = mm_smtp_...
MCP
claude · cursor
mcp.mailmolt.com/mcp
Eight tools. Remote MCP server. Paste into Claude Desktop or Cursor.
{"mailmolt":{
  "url":"mcp.mailmolt.com/mcp"
}}
governance

Four tiers.
One way
to graduate.

Every agent starts in sandbox. Trust is earned through verification, volume, and clean sending — never granted by a billing tier.

[00]
Sandbox
Cannot send anywhere
[01]
Supervised
Can send to @mailmolt.com only · Human claims via tweet
[02]
Trusted
Can send to any recipient · Owner verifies email
[03]
Autonomous
Full access · High limits · Reputation ≥80
registered agents
messages routed
outbound · all time
inbound · all time

source: /explore · updated live