Manage an email platform from your AI assistant: Lettr's 79 MCP tools
An AI assistant such as Claude, ChatGPT, or Cursor can operate Lettr through 79 MCP tools, from sending and scheduling to campaigns, audience management, and delivery diagnostics, with more added as Lettr ships new features. This article covers what the tools do, the two ways to connect, and the guardrails that limit what an assistant can send or change.
Lettr hosts a remote MCP server at app.lettr.com/mcp that AI assistants such as
Claude, ChatGPT, and Cursor connect to. It has 79 tools across ten areas of the product, up from
18 before the September 2026 release, and covers most day-to-day work in Lettr: sending,
scheduling, campaigns, the audience, and delivery diagnostics. The count keeps
rising, because new tools ship alongside new Lettr features.
What the Lettr MCP server covers
Every tool with a REST counterpart reuses that endpoint's action, so a tool call reads and writes the same data an API client would. The ten areas are emails, scheduled emails, templates with their folders and projects, sending domains, the other domain types (tracking, inbound, and storage), webhooks, campaigns, the audience, analytics and monitoring, and the account itself.
When we wrote about the MCP integration
in May 2026, the remote server had 16 tools, mostly for reading account data, plus sending
email and creating and updating templates. The September release brought it to parity with the lettr-mcp npm package, adding campaigns, the
audience, scheduling, domains, and webhooks. Once the remote server also accepted API keys, the
one thing the npm package offered that a sign-in could not, the package was deprecated.
What an assistant can do with the full set
The largest additions are campaigns and the audience. A question like "How did last week's newsletter campaign perform?" returns the campaign's engagement stats, and "create an opt-in topic called Product updates" creates the topic. Campaigns are still built in the dashboard; the assistant sends, schedules, and reports on them.
The remote server also carries four diagnostic tools the npm package never had: DNS diagnosis, a sending health check, analytics, and API logs. Together with each email's delivery timeline, they answer questions such as why a message failed to deliver, or whether a domain is verified and ready to send, in the same conversation that sent the email.
Two ways to connect: sign in or use an API key
Signing in connects the server through OAuth, and every tool then acts as the
signed-in user, in the team they choose. Each call goes through the same team permission checks
as the dashboard, and two account tools, list_teams and current_team,
show which teams the connection can reach and which one a call would act on. The sign-in flow works from Claude.ai, Claude Desktop, Claude Code, ChatGPT,
Cursor, and GitHub Copilot, and it is the right choice for interactive use.
An API key suits anything unattended, such as CI jobs, cron scripts, and headless agents, because it needs no browser and no sign-in. The client sends the key as a bearer header, and the key's scopes decide which tools the assistant can use. In Claude Code, that is one command:
claude mcp add --transport http lettr https://app.lettr.com/mcp \
--header "Authorization: Bearer lttr_xxxxxxxxx"The same server can be connected twice, signed in for chat and with a scoped key for automation, under different names so their tools stay easy to tell apart.
What stops an assistant from sending email by mistake?
Four guardrails limit what an assistant can do. The first applies to every connection, and the other three apply to connections made with an API key.
- Destructive annotations. Tools that send email right away, overwrite data, or delete it carry the MCP destructive annotation, so a client such as Claude can ask for confirmation before calling one.
- Scopes. When nobody is present to confirm, the key's scopes do the restricting. The server does not list a tool the scopes exclude, so a headless agent cannot call it.
- Sandbox keys. Every email sent with a sandbox key has its sender and recipient rewritten, so an agent holding one cannot email a real customer, whatever its prompt says.
- API logs. Every call made with an API key is recorded in the API logs with the key's ID, so an agent's actions can be reviewed afterwards. Calls made by signing in belong to a person rather than a key and don't appear there.
The remote server also runs in its own Lambda function, so the functions serving the web app and the REST API have no access to the OAuth signing keys.
An AI client sends everything a tool returns to its model provider, contact details included, and that part is outside Lettr's control, so check the provider's data terms before managing the audience from an assistant. Connect only AI clients you trust, and review an action before confirming anything sensitive.
FAQ
Which AI assistants can connect to Lettr's MCP server?
Can the remote MCP server run campaigns and change the audience?
Can I limit what an agent is allowed to do?
Do I still need the lettr-mcp npm package?
Will Lettr add more MCP tools?
Bottom line
An assistant can now send campaigns and manage the audience, on top of the sending, templates, and diagnostics it already handled. Sign in for chat, and give anything unattended a scoped key so its scopes and the API logs keep it in check.
Create a free Lettr account to connect an assistant; the free plan needs no credit card, and the MCP server works on it. Once you're in, the MCP docs walk through each client.