Claude calls the tool. Senddio sends it.
MCP runs the server as a local stdio process. Claude asks the model to invoke a tool, the server forwards an authenticated POST to /v1/send, the response flows back. Your agent key controls everything — channel scope, IP allowlist, expiry.
- Local stdio process — no network ingress on your side
- Strongly-typed tool schemas with Zod-style validation
- Auto-retries on 5xx with exponential backoff
- Agent-name baked into every send for clean audit
- Works with Claude Desktop, Claude Code, and the SDK
User → "Ping #ops that build #2974 shipped."
Claude → calls send_message({
channel: "slack",
to: ["C0123456789"],
body: "build #2974 → prod ✓"
})
← { id: "sdo_msg_…", status: "queued" }
Claude → "Queued. Check message status for delivery.
Logs tagged agent=claude."Five tools, one envelope.
Your first 30 days with Claude on the wire.
Claude MCP — FAQ
MCP, the Model Context Protocol, is Anthropic's open standard for letting Claude desktop and IDE clients use external tools safely. Senddio ships a first-class MCP server so any Claude instance can send a message, check a status, or read inbound replies the moment you connect the config.
Yes, but it stays separate from yours. You create an agent-scoped key in the dashboard — with the channels it can use, the IP addresses it can call from, and an expiry date — and add it to the MCP server's environment. Every message Claude sends is labelled with the agent name, audited, and rate-limited on its own.
Yes. The server is published on npm as @senddio/mcp-server under a permissive licence, with the source available on GitHub. Review release notes and test your client configuration before upgrading.
Yes. MCP is a wire protocol, so any MCP-compatible client can load the Senddio server — Clients supporting local stdio MCP servers can load it. Hosted clients need an appropriate remote bridge. For ChatGPT and other OpenAI assistants, use our OpenAI tools package instead.
Agent sends count against your plan's allowance just like any human send — there is no premium for being an agent. Your dashboard splits agent traffic from human traffic so you can see exactly what your AI is doing and what it costs.