Toolbox
Ready-made pieces you can extend: Outbox and HITL policies, mail and calendar, documents with RAG, chat, and user management—wired for MCP clients.
Ironbrain
Critical MCP actions land in the Outbox. You decide which tools need human approval—and which may run without interrupting the flow.
Confirm, edit, or reject before execution
Reply with revised proposal timeline
Hi Anna — thanks for the call. Attached is the revised timeline…
Agent moved the workshop · review before / after
Ran automatically · your policy
Every connected agent hits one MCP endpoint. Your HITL policy decides: hold in the Outbox, or run on the auto path.
The Outbox mirrors what Ironbrain already ships: typed actions, editable payloads, before/after diffs, Approve / Edit / Reject, and a hard floor for external sends.
Approve action
Approve action
Agent moved the workshop. Review before / after, then approve.
Every action has an HITL mode. It starts from a default and can be overridden for your organization and in your personal settings. Critical tools wait in the Outbox; the rest stay automatic.
Always wait in the Outbox
Ask when below threshold
Run unless policy intervenes
No approval (usually reads)
Approval is the core product. Around it we offer building blocks, custom integrations, and advice for bringing AI into your stack.
Ready-made pieces you can extend: Outbox and HITL policies, mail and calendar, documents with RAG, chat, and user management—wired for MCP clients.
Individual solutions with deeper vertical integration for specific workflows—delivered by us or with partners, fitted to your processes.
AI integration consulting and onboarding: we help you connect assistants to real systems safely—with Ironbrain or alongside other services.
Interested in the platform, a custom solution, or consulting? Write to us.
hello@ironbrain.de