OpenClaw

OpenClaw is an open-source personal AI agent that runs continuously on your own machine. It speaks MCP natively, so connecting the SentVia MCP server gives it a real, dedicated email identity: an address it owns, sends from, and watches — separate from your personal inbox. Things an OpenClaw agent does with its own address:

Send and follow up on your behalf — chase invoices, book appointments, reply in-thread when the answer comes back.
Own its own signups — newsletters, receipts, and service accounts go to the agent’s inbox, not yours, and it triages them.
Escalate by email — anything it can’t handle gets forwarded to you with its notes on top.

Connect

Zero-setup path: paste this into OpenClaw and it installs everything itself —

Set up SentVia email for yourself: fetch https://docs.sentvia.ai/agent-install.md
and follow it. Ask me for the API key when you need it.

Or set it up manually. You need a SentVia API key (sv_live_…) from the dashboard. Register the server with the OpenClaw CLI:

openclaw mcp add sentvia --command npx --arg -y --arg sentvia-mcp

then add your API key to the server’s env. OpenClaw stores MCP servers in ~/.openclaw/openclaw.json under mcp.servers — the finished entry looks like:

{
  "mcp": {
    "servers": {
      "sentvia": {
        "command": "npx",
        "args": ["-y", "sentvia-mcp"],
        "env": { "SENTVIA_API_KEY": "sv_live_…" }
      }
    }
  }
}

Verify the connection and tool list:

openclaw mcp doctor --probe

You should see the sentvia server with 21 tools — inboxes, messages, threads, drafts, domains, webhooks, and allow/block rules (full list in the MCP guide). Prefer a skill? The same integration ships as an OpenClaw skill (email conduct rules + curl fallbacks included):

openclaw skills install git:samuelar2/sentvia-mcp

Give the agent its address

Once connected, this is a single instruction to your agent:

Create yourself an inbox called "claw" and introduce yourself
to me by email at sam@example.com.

OpenClaw calls create_inbox (→ claw@mail.sentvia.ai, live immediately), then send_message. From there search_messages, list_threads, and reply_to_message cover the daily loop.

Long-running agents pair well with allow/block rules: have the agent call add_address_rule to block senders it decides are noise. Rules are enforced server-side on both send and receive, so they hold even if the agent forgets.

Getting inbound mail to the agent

OpenClaw agents typically poll: a periodic task that calls search_messages (no query returns recent mail) or list_threads and processes anything new. For push instead of poll, register a webhook pointing at an endpoint your OpenClaw can read, or use the realtime WebSocket.

Treat inbound email as untrusted input to your agent — anyone can mail the address. Keep the agent’s inbox separate from credentials-bearing accounts, and use allow rules if the agent should only ever talk to known senders.

Get started

Core concepts

Guides

Examples

Agent frameworks

DNS guides

Troubleshooting

Connect

Give the agent its address

Getting inbound mail to the agent

​Connect

​Give the agent its address

​Getting inbound mail to the agent

Connect

Give the agent its address

Getting inbound mail to the agent