Give Lovable an email inbox
Add Anjal as a custom MCP server in Lovable's Connectors. Chat connectors are on by default on every plan, so it is one URL and an OAuth click.
Lovable generates the app. Almost every app it generates eventually needs email, and "eventually" is usually the moment you want a signup confirmation, a password reset or a contact form that actually reaches someone.
Anjal is a custom MCP server you add once as a chat connector.
Connect it
Open Connectors, go to the Chat connectors tab, and click New MCP server.
- Name:
Anjal - MCP endpoint URL:
https://mcp.anjal.app/mcp - Authentication: leave it on OAuth, which is Lovable's default when the server supports it. Ours does.
Lovable registers itself, sends you to Anjal to sign in and click Allow, and the tools appear. There is no API key to paste, which matters more here than in an editor: a key pasted into a generated app has a habit of ending up in the repo.
Chat connectors are enabled by default on every plan, including Enterprise. The only Enterprise-gated part is Lovable's static-IP option for custom MCP servers, which you do not need for us: we do not require IP allowlisting.
What Lovable gets
47 tools. Building an app, these are the ones that come up:
| Tool | What it does |
|---|---|
create_inbox |
An address for the app, on your domain or ours |
send_message |
The confirmation, the reset, the receipt |
create_webhook |
Fire the app's endpoint when mail arrives |
list_messages, get_thread |
Read replies, in thread order |
search_messages |
Find the message you are testing against |
reply_to_message |
Reply in-thread, headers written correctly |
verify_email |
MX pre-flight on an address before you rely on it |
Why this is worth doing at generation time
The generated app gets working email instead of a TODO. Lovable can create the inbox, register the webhook and wire the send call while it is building, so the app that comes out the other side already sends and receives. That is the difference between a prototype and something you can put in front of a user.
Email is a hard thing to add later. SPF, DKIM, DMARC, a sending domain with a reputation, bounce handling. Starting from an inbox that already has all of that is a much better position than bolting an ESP on after launch and discovering your mail goes to spam.
A chat connector is also a research tool. Lovable can look things up in your Anjal account while it builds, so "what does the payload of an inbound message look like" is answered from your real data rather than guessed.
Scoping the app, not just the connector
The connector is for Lovable itself. For the deployed app, use a key rather than MCP, and scope it:
curl -X POST https://anjal.app/v1/inboxes/$INBOX/send \
-H "X-API-Key: $ANJAL_KEY" -H "Content-Type: application/json" \
-d '{"to":"user@example.com","subject":"Confirm your email","plainBody":"..."}'
Use a workspace-scoped wk_ key in the app's environment, never your tenant master key. It reaches one workspace and cannot see that any others exist, so a key that leaks out of a generated app costs you that workspace and nothing more. Your plan gives you one workspace per inbox, so an app per workspace is usually affordable.
Before the app reads mail from strangers
Anything a user emails your app is attacker-controlled text, and if you feed it to a model it is an injection surface. Anjal screens every inbound message on the ingest path and quarantines what it judges hostile before an agent can read it; an agent cannot release its own quarantine. We publish the architecture and not the rules. Agent safety is honest about the limits: detection is advisory, and gating side-effects on the sender in code is the part that actually holds.
Cost
Starter is $5/month for 2 inboxes and 6,000 emails. Developer is $15/month for 10 inboxes and 20,000 emails. Extra inboxes are $2/month each. 7-day trial.
Next
Setup verified against Lovable's MCP servers documentation on 2026-09-28.
Other clients
- One command, then sign inClaude CodeConnect Anjal to Claude Code with one command. Your coding agent gets a real inbox it can read, reply from and thread, plus a calendar, over MCP.
- Add, then log inCodexConnect Anjal to OpenAI Codex with codex mcp add and a login. Your agent gets a real inbox it can read, reply from and thread, plus a calendar.
- One JSON entryCursorConnect Anjal to Cursor by adding one entry to mcp.json. Your editor's agent gets a real inbox it can read, reply from and thread, plus a calendar.
- Org admin adds it onceDevinAdd Anjal to Devin from the MCPs tab or with devin mcp add. An autonomous engineer that can receive mail can report in, and get itself past a signup wall.
- One TypeScript fileOpenClawOpenClaw takes a TypeScript extension file rather than an MCP server. Drop one file in ~/.openclaw/extensions/ and every agent gets native Anjal tools.
- Two clicks, no fileReplitAdd Anjal as an MCP server in Replit's Integrations page. Two clicks, no config file, and it works across every project on your account.
Setup last verified 2026-09-28. Clients change their config formats; if a command here is wrong, tell us at kamal@send.anjal.app.