MCP (AI Access)
Sendbox exposes a remote MCP server so your own LLM — Claude, or any other MCP-capable client — can operate your workspace directly: list and add leads, build and run campaigns, work the inbox, check analytics, and more, all as tool calls the model makes itself.
https://mcp.sendboxes.tech/mcpThis is not a separate product — it’s the same REST API your dashboard uses, exposed as MCP tools. Anything a tool does, it does through the real API route for that action.
Setup
Get an API key
Go to Settings → API Keys in the dashboard and create one. It’s shown in full exactly once — copy it now.
Only an API key (sb_live_...) works here — a dashboard login session does not. This is
deliberate: an API key is scoped to one workspace and already can’t touch billing, team, or
workspace settings, so a persistent LLM connection is exactly what it’s designed for.
Add the connector to your MCP client
The server authenticates with your key in an Authorization: Bearer sb_live_... header.
Claude Code
Run:
claude mcp add --transport http sendbox https://mcp.sendboxes.tech/mcp --header "Authorization: Bearer sb_live_..."Or add it to your project’s .mcp.json:
{
"mcpServers": {
"sendbox": {
"type": "http",
"url": "https://mcp.sendboxes.tech/mcp",
"headers": {
"Authorization": "Bearer sb_live_..."
}
}
}
}Create the key as Read only if the AI should only look, never change or send anything. It will then only be offered the read-only tools. Each key is limited to 300 requests per minute, and every tool call counts against it.
Start asking
Once connected, you can ask your LLM to do things like:
- “List my active campaigns and their reply rates.”
- “Add these 20 leads and put them in the ‘Q3 Outreach’ campaign.”
- “Show me every reply marked Interested this week.”
- “Draft a reply to the latest message from [email protected].”
What it can and can’t do
See Available Tools for the full list, and Security Model for what’s deliberately excluded and why.