AI agents and MCP
Mailpark is built to be set up by your coding agent. Connect it over MCP and ask it to set up your emails: it adds templates, wires sends into your code, tests them and prepares your domain.
Connect the MCP server
The server speaks streamable HTTP at https://api.mailpark.io/mcp. Give it a test key, so nothing the agent does reaches real people.
claude mcp add --transport http mailpark https://api.mailpark.io/mcp \
--header "Authorization: Bearer sk_test_…"Then ask, for example:
Set up Mailpark for this app: add branded templates, send the password reset and
welcome emails from our auth code, and test them in the test inbox.Tools
Each result is a short summary, the JSON, and "Next steps". Errors include a Fix line. Start with get_started: it returns the setup checklist and the next step.
llms.txt
/llms.txt is the whole API in one plain-text file, written for agents: auth, every endpoint, one complete example per job, the safety rules and the error format. Point agents without MCP at it.
Guardrails
Agents can draft anything, but some things always wait for a person:
- Broadcasts wait for approval. The agent gets an
approval_urlto hand over. - Going live, billing, spending caps, API keys and the team are web-app only.
- Live keys can't turn sequences on:
sequences:activateis test-only. - Every action is in the audit log under the key's name.