Connect Claude Code

How to connect Claude Code to email (send and reply via MCP)

Connect Claude Code to email by adding SendRaven's remote MCP server with one command: claude mcp add --transport http sendraven https://mcp.sendraven.ai/mcp.

Claude Code can then send from your own domain, list the conversations waiting on a reply and answer inside the same thread, within the limits of the API key it connected with.

Claude CodeTerminal
claude mcp add --transport http sendraven https://mcp.sendraven.ai/mcp
Then ask
> Which email threads are waiting on a reply? Draft an answer to each and hold them for me to approve.

Before you start

What Claude Code needs.

Claude Code installed and signed in. Any plan that runs Claude Code is enough; the MCP server costs nothing extra.

A SendRaven workspace with one verified sending domain, and a payment method on it. Free includes 3,000 emails a month and is never charged without an upgrade.

For an agent that runs without you watching, an API key with the limits you want. A browser sign-in is fine for working alongside it.

Setup

Connect Claude Code step by step.

  1. Add the server

    Run this in the project you want Claude Code to email from. It is added for that directory only, which is Claude Code's default scope; add -s user to have it everywhere.

    Terminal
    claude mcp add --transport http sendraven https://mcp.sendraven.ai/mcp
  2. Sign in, or pass a key

    The first tool call opens a browser to sign in and pick a workspace. For an unattended agent, add the server with a key instead, so the key's daily cap, allowlist and approval hold apply: a browser sign-in carries none of them.

    Terminal, with an API key
    claude mcp add --transport http sendraven https://mcp.sendraven.ai/mcp \
      --header "Authorization: Bearer sk_live_…"
  3. Check it connected

    claude mcp list shows the server with a tick when Claude Code can reach it. Inside a session, /mcp shows the same and lists the tools.

    Terminal
    claude mcp list
    sendraven: https://mcp.sendraven.ai/mcp (HTTP) - ✔ Connected
  4. Ask for something

    Ask in plain words. Claude Code picks the tools: get_usage before a batch, list_threads to see what is waiting, send_email or reply_to_message to answer.

Try it

Three things to ask Claude Code.

“Which email threads are waiting on a reply? Draft an answer to each and hold them for me to approve.”

Runs list_threads with awaiting_reply, reads each with get_thread, and replies with reply_to_message. On a key with an approval hold every reply comes back pending_approval for you to release.

“When the deploy finishes, email the release notes in CHANGELOG.md to beta@yourcompany.com from releases@mail.yourcompany.com.”

Reads the file from the repository it is already working in and sends it with send_email. This is where Claude Code differs from a chat app: the mail can come from the work it just did.

“How many emails have we sent this month, and how many are left before the plan limit?”

Calls get_usage, which reports the plan, what has been sent and what remains, so a batch can be sized before it is sent rather than refused halfway.

Once connected

What Claude Code can do with email.

Send

From your own verified domain, as plain text, HTML or a stored template, now or at a set time.

  • send_email
  • send_template_email
  • list_scheduled_emails

Read the replies

Replies join the thread they answer on Message-ID, with the quoted history removed, and one call lists what is waiting on you.

  • list_threads
  • get_thread
  • mark_thread_handled

Reply in the thread

The answer goes into the same conversation with the right headers, so it lands under the original in the recipient's inbox.

  • reply_to_message

Run sequences and campaigns

Automations that stop when someone replies or unsubscribes, and campaigns that can be previewed before they go.

  • create_automation
  • enroll_in_automation
  • preview_broadcast

Every tool the server has, grouped by what an agent is trying to do.

Safety controls

The limits live on the key, not in the prompt.

Claude Code reaches exactly what its credential can reach. Give an agent's API key any of these, and the tools respect them because the API underneath enforces them.

A daily send cap

daily_send_limit stops the key once it has sent that many today, counting scheduled and held mail too.

A recipient allowlist

allowed_recipients takes addresses and whole domains. Anything else is refused before it is written.

An approval hold

requires_approval turns every send into a draft that a person releases. The key cannot approve its own drafts.

A browser sign-in carries none of the three, so use a key for anything that runs unattended. A key also cannot raise its own limits: key management has no MCP tool. Every limit a key can carry.

Troubleshooting

What went wrong when we tested it.

A tool answers "This key is missing the domains:read scope (HTTP 403 forbidden)".

The key works, it just was not given that scope. Either add the scope to the key in the dashboard, or tell Claude Code which address to send from so it does not need to look the domains up. A key made for sending and reading threads only is a reasonable choice for an agent.

send_email comes back with status pending_approval and nothing arrives.

That is the approval hold doing its job, not a failure. The message waits under Approvals in the dashboard until a person releases it. Asking Claude Code to try again queues a second draft, so do not.

The server is missing from claude mcp list in another project.

claude mcp add registers the server for the current directory unless told otherwise. Run it again with -s user to make it available in every project, or with -s project to write it to .mcp.json and share it with the repository.

Tested end to end on 15 November 2026.

Why not just Resend?

Resend has an MCP server too. The difference is what happens after the send.

The reply is a thread, not a webhook

SendRaven: Joined on Message-ID to the message it answers. Resend: a received-email event; you fetch the message and set reply headers yourself.

Per-key daily cap and recipient allowlist

SendRaven: daily_send_limit and allowed_recipients on the key, checked on every send. Resend: keys scope to a domain and to send-only; no cap or allowlist.

Human approval hold on a key

SendRaven: requires_approval drafts the message and waits for a person. Resend: not offered.

From resend.com, last checked 16 September 2026. The full comparison, including where Resend is the better choice.

Questions

Can Claude Code send email?

Yes, through an MCP server that wraps an email API. With SendRaven connected, Claude Code calls send_email to send from your own verified domain, and can also read replies, answer inside the thread, schedule sends and run campaigns.

Can Claude Code read the replies to email it sent?

Yes. Replies arrive as threads joined on Message-ID, list_threads with awaiting_reply returns the conversations waiting on an answer, and get_thread returns each one with the quoted history removed, so Claude Code reads what the person actually wrote.

How do I stop Claude Code from sending email without asking me?

Connect it with an API key that has an approval hold. Every send then returns pending_approval and waits for a person under Approvals, and the key cannot approve its own drafts or change its own limits. A daily cap and a recipient allowlist can sit on the same key.

Is there a Claude Code plugin for SendRaven?

Yes. /plugin marketplace add CommonNinja/sendraven-mcp-server, then /plugin install sendraven@sendraven, installs the same remote server. It signs in through the browser on first use, like the one-line install.

Give Claude Code an inbox it cannot misuse.

Verify a domain, create a key with the limits you want, and connect. Free includes 3,000 emails a month, inbound included, and is never charged without an upgrade.