What Mittr is

Mittr is the reliable action layer. It delivers every action your software takes, whether your code fires it (webhooks) or your AI decides it (agent actions), with signing, retries, dead-letter, replay, and a full audit trail. Then it proves the action happened.

An outbound action is an outbound action. A Stripe-style order.created webhook, a “post to Slack” call, and an AI agent firing “issue refund” are the same act: deliver a signed, retried request to an endpoint, and record exactly what happened. Mittr models that act with four primitives:

• Action (an event): the thing to deliver, with a JSON payload and an idempotency key. Created by your code via the REST API, or by an agent via the MCP tool surface.
• Endpoint: a destination URL that receives the action. Private and internal URLs are rejected (SSRF protection).
• Delivery attempt: each try at reaching the endpoint, with status code, latency, and error. Failed attempts are retried with exponential backoff and eventually dead-lettered.
• Audit trail: the durable record of the action and every attempt, so you can prove what was dispatched, what landed, and what was retried or replayed.

Mittr has two entry points into the same delivery engine and the same audit trail:

• Code-triggered (webhooks). Your systems fire events; Mittr fans them out to subscribed endpoints. This is the proven path, battle-tested on production traffic. See the Quickstart.
• AI-decided (agent actions). Your agent dispatches actions over the Model Context Protocol; A2A push notifications (authenticated webhooks between agents) arrive inbound. The action lands and is provable even when the agent is not.

Inbound, other systems and agents deliver actions to you, verified against 135+ platform presets. See Inbound Webhooks.

Whether the trigger is code or an AI agent, the hard part is the same: making the action reliably land and be verifiable. Agents make this urgent, they decide well but execute unreliably (a tool call can be formed correctly and still fail, and a naive retry can fire the same side effect twice). Mittr is the layer that closes that gap: it delivers the action, retries it safely via idempotency keys, dead- letters what cannot be delivered, and records the whole lifecycle so you can prove and replay it.

Next: the reliability and audit guarantees that make this work.