Cursor's agent panels build features end to end. Ask them to add a signup email or a contact form and, if the tools are there, they finish the job. mailnix gives the agent something to reach for: a hosted MCP server it can wire in one config edit and use from every project on the machine.
01
Add mailnix once in the global Cursor settings or per-project in .cursor/mcp.json. Cursor's Compose and Agent surfaces read the tool catalog on the next reconnect.
02
Cursor's private-use URI callback (cursor://anysphere.cursor-mcp/oauth/callback) is accepted at /oauth/register, with a loopback fallback on 127.0.0.1. Consent runs in a browser tab; the token lives with the Cursor client.
03
Cursor re-registers via DCR on every reconnect. mailnix mints a fresh client_id per registration and shares no per-client state, so the reconnect loop never wedges the server.
04
cursor.com and app.cursor.sh are in the CORS allowlist, so Cursor's browser surfaces reach the same server without a separate config.
mcp.json entry
{
"mcpServers": {
"mailnix": {
"url": "https://mcp.mailnix.ch/mcp"
}
}
}form_create builds the form. email_provider_connect wires the sender. email_send_test fires a message. email_trace_lookup returns the event list Cursor can paste back into the diff explanation.
form_create, form_publish, form_embed_snippet_generate return a copy-paste embed. The agent can drop it into the JSX and update the routing rules in the same session.
When a customer says the receipt did not arrive, paste the trace ID and ask Cursor to run email_trace_lookup. The event list comes back inline, no dashboard tab required.
publishable_key_create hands back a five-minute reveal URL, not the raw token. The agent can wire origin lock, recipient allow-lists, and rate caps in the same call.
OAuth 2.1 with PKCE S256 (mandatory), Dynamic Client Registration, and audience-bound JWTs. Private-use redirect URIs (RFC 8252 section 7.1) and loopback fallbacks are accepted, so both the desktop and web surfaces of Cursor complete the flow the first time.
Yes. Both read the MCP server list from your Cursor settings, so mailnix tools show up in either surface once the connection is authorized.
No. On the first connect, the anonymous sandbox bootstrap gives Cursor a working token against a fresh sandbox project. Claim it into an account later without editing mcp.json.
Cursor re-registers via DCR each time. mailnix issues a new client_id and reuses no client-side state, so the flow is stable across restarts and updates.
Yes. The OAuth token is scoped (mcp:read, mcp:send-sandbox, mcp:send, mcp:admin). Grant a narrower set at the consent screen and the server intersect-downgrades to those scopes for every subsequent tool call.
In EU data centres in Germany, run by LIMERION GmbH out of Liechtenstein. GDPR-compliant. Signed DPA on request.
One HTTP call to get started