What is Protobox?
AI that knows your business, follows your playbook, and does the work — through one managed MCP endpoint.
Protobox is a managed backend for your AI. You connect your company's apps and content once; every AI assistant your team uses — Claude, Cursor, or any MCP-compatible client — gets the same equipped, controlled access through a single endpoint:
https://your-workspace.protobox.app/mcpNo servers to run, no OAuth plumbing to build, no RAG pipeline to maintain.
The problem it replaces
Without a managed layer, giving AI real access to your stack means workarounds: API keys pasted into chat windows, a human copying answers between tools, brittle one-off automations, and a different copilot bolted onto every app. Each workaround is a security hole, a maintenance burden, or both.
Equipped
Your AI does real work because the endpoint carries everything it needs:
- Tools — actions in your connected apps (Slack, HubSpot, Google Workspace, and more) plus custom HTTP and JavaScript tools you define.
- Knowledge — your documents, URLs, and files, searchable so answers come from your content instead of guesses.
- Prompts — your playbooks, exposed as reusable prompts clients can invoke.
MCP defines three primitives — tools, resources, prompts. Protobox hosts all three.
Controlled
Access is yours to shape. Everything lives in a workspace you administer: connections are approved by you, API keys are workspace-scoped and revocable, and toolsets let each client see only the tools it needs — support agents don't get deploy tools.
Everywhere
Because Protobox speaks MCP — the open protocol AI clients standardize on — the same workspace plugs into Claude Desktop, Claude Code, Cursor, and any other MCP client, today and as new ones appear. Set up once, use everywhere.