MailCheap: Open-Source Newsletter App You Can Run From Claude or ChatGPT

A newsletter tool is trending on GitHub because of its pricing model, not because of AI. MailCheap from Marc Lou, at marclou/mailcheap, has about 1.3k stars with a simple pitch from its README: pay for the emails you send, not for dormant subscribers. Tucked into the feature list is the reason it belongs on an AI blog: every dashboard action is also exposed through an MCP server (the open standard for connecting AI applications to external systems) and a REST API, so Claude, ChatGPT and Grok Bot can run your newsletter.

If you are choosing a newsletter stack, or building agents that touch email, this is a useful case study in how a small product handles the risk of letting an agent press "send".

TL;DR: questions people ask first

Why the pricing model matters

Most newsletter platforms charge by subscriber count, which means a list full of people who stopped opening years ago still costs you every month. MailCheap bills by volume sent. For a list with many dormant addresses, the arithmetic can favour sending-based pricing. The catch is that your cost scales with how often you send, not how big your audience is, so daily senders see a different curve than monthly ones.

The self-hosted route is cheaper still: you pay only your email provider. On Amazon SES the README quotes roughly $0.10 to $0.16 per 1,000 emails, which for a 20,000-email send is a couple of dollars. Compare that with the $19 hosted entry plan and the trade-off is clear: money for convenience and deliverability work versus your own time.

What is in the box

The feature list in the README covers the whole newsletter workflow:

• Writing, review checks and sending, scheduling, or staged sends to engaged readers first.
• CSV imports from beehiiv, Substack, Mailchimp and several other platforms.
• A public website per newsletter with signup forms and a sitemap.
• Sending from your own domain, with DKIM, return-path and DMARC checks.
• Deliverability protections: warm-up, list cleanup, bounce and complaint limits, one-click unsubscribe, and optional Google Postmaster Tools monitoring.
• Open, click and subscriber stats.

Deliverability tooling is the real engineering here. Warm-up and complaint limits are what stop a new sending domain from landing in spam, and they are exactly the part self-hosters tend to get wrong.

The AI angle: an MCP server for your newsletter

According to the README, dashboard actions are available through an MCP server and a REST API, "with the same checks and limits". ChatGPT, Claude and Grok Bot connect through OAuth, while Claude Code, Codex and Cursor use an API key. If you are new to the protocol, our MCP explainer and the list of open and closed MCP servers give the background.

In practice that means you could ask an agent to draft this week's issue from your notes, run the review checks, schedule it for Tuesday, and report the stats on Friday, all without opening the dashboard. Other tools are moving the same way: DaVinci Resolve 21.1 added an MCP server so agents can edit video, and Grok's agent is reaching into more apps, as in our Grok Bot coverage.

How it treats irreversible actions

The detail worth copying is that sending is flagged as an action that "can't be undone", and agents are designed to ask for confirmation first. An email blast cannot be recalled, so the design puts a human approval at the one step that matters. That is the same principle behind the ChatGPT Dots unapproved email incident, where an agent emailed without sign-off. We cannot verify how strictly the confirmation is enforced; the README states the intent, and you should test it yourself with a small list before connecting an agent to a real audience.

Self-hosting stack and gotchas

Known limits from the README:

• Self-hosting has no support, issues are turned off, and pull requests may not be reviewed.
• Sending requires provider approval. SES starts in a sandbox, and PostShiba approves production sending by hand.
• Without Redis, rate limits are not shared across server instances. Without S3 or Blob, image uploads are off.
• Some features, such as DNS automation for shared addresses, need a separate DNS helper that is not in this repository.
• MailCheap Cloud runs the same code plus billing, but its marketing site is not open source.

AGPL-3.0 also deserves a note: if you modify the software and offer it to others over a network, you must publish your changes. That is fine for running your own newsletter and a real constraint if you plan to resell a modified version.

Worked cost examples

These use only the figures in the README, so treat them as rough estimates, not quotes.

The self-hosted figures leave out your hosting, MongoDB and QStash costs, which depend on volume and provider. The point is the shape of the curve: with per-email pricing, a list that sends rarely is cheap whatever its size, and a list that sends weekly pays for every send. Subscriber-priced tools invert that, so the right choice depends on how often you publish and how much of your list is inactive.

Migrating from another platform

The README lists CSV imports from beehiiv, Substack, Mailchimp and several other platforms. A sensible migration plan, based on that feature set:

1. Export subscribers from your current tool and remove addresses that have not engaged for a long time. The deliverability protections include list cleanup, but a clean import protects a new sending domain's reputation.
2. Set up your sending domain with the DKIM, return-path and DMARC checks the app provides.
3. Request production sending from your provider. On Amazon SES, accounts start in a sandbox, so allow time for approval.
4. Warm up gradually instead of sending your full list on day one; the app's warm-up limits exist for this reason.
5. Use staged sends to engaged readers first, which the README lists as a feature, to watch bounce and complaint rates before the rest of the list.

What this says about agent-ready products

MailCheap is a small example of a larger shift: products now ship an MCP server and OAuth connectors alongside the dashboard, because users expect to run them from an assistant. The design questions are the ones every such product faces. Which actions need confirmation? Do agents and humans share the same rate limits? Can a leaked API key send mail to your whole list? Here the README answers the first two directly, with the same checks and limits for both and a confirmation step for irreversible sends. It does not describe scoping for API keys, so assume a key grants broad access and store it accordingly.

Should you use it?

Use it if you are a creator with a mixed-quality list and send occasionally, if you want to own your sending infrastructure, or if you want agents to handle the repetitive parts of publishing. Stay on a managed platform if you want vendor support, more than three newsletters per account, or built-in monetization tools the README does not mention. Our list of AI newsletters to follow shows how many are still on subscriber-priced tools.

Safety checklist before connecting an agent

1. Create a test newsletter with a handful of your own addresses.
2. Connect the agent with the narrowest credentials available; prefer OAuth where offered and rotate API keys.
3. Ask it to send, and confirm it stops to ask first.
4. Check that unsubscribe links and complaint handling work as described.
5. Log what the agent did. For stronger guardrails around agent actions, AgentBeam is the agent security platform from the explainx.ai team and stops agents before they take dangerous actions.

Related reading

• What is MCP? The complete guide
• Top 10 open and closed source MCP servers
• Top 10 AI newsletters to follow
• ChatGPT Dots and the unapproved email
• Grok Bot bank balance and Slack permissions
• DaVinci Resolve 21.1 MCP server

Pricing, features and star counts are taken from the MailCheap repository as of October 23, 2026 and may change.