Mumla is the approval-and-proof layer for AI communication.
When AI proposes a message in your company's name, Mumla routes the decision to the right person, records exactly what they approved, sends only what was authorized, and keeps tamper-evident proof.
AI does the work. People make the calls. Mumla keeps the proof.
How it works
A governed gateway. Three concepts, end to end.
Mumla is a governed communication gateway for AI agents — exposed first as an MCP server. The agent's config holds a Mumla URL instead of Gmail, Slack, or Twilio credentials. Mumla holds the real credentials; the agent gets capabilities, not keys.
- 01
Request action
The agent asks to send, contact, ask, or escalate. It never touches your channels directly.
- 02
Resolve authority
Mumla decides: allowed, pre-approved, or needs a human. Authority rides on scoped grants — per agent, per recipient, per purpose.
- 03
Return receipt
Tamper-evident proof of what happened and why — signed, retained, and verifiable in your own browser.
The agent gets capabilities, not credentials.
The human gets decisions, not dashboards.
The company gets proof, not archaeology.
Capability 01
Agent ↔ human communication
Ask a person and durably wait. Escalate through a chain. Gate an outbound send on approval. Notify — governed, never raw.
Read more →
Capability 02
Decision cards where you work
No new inbox. The decision arrives as a rich card in email or Slack, with everything needed to approve or reject — from your phone if you like.
Read more →
Works with your agents
Anything that speaks MCP plugs in with one URL and one scoped key.
Delivers where you work
Seven channels, built and tested — the hosted gateway opens with email. Rich decision cards land on email and Slack first.
Don't trust us — verify.
Every receipt in the ledger carries a real ed25519 signature. Open one, verify it in your own browser, then flip a single bit and watch it fail.