emiliaprotocol/emilia-protocol

npm install @emilia-protocol/verify  (JS) / pip install langchain-emilia  (Python)
Run the local surface map: npx @emilia-protocol/scan protect ./tools.json  then  npx @emilia-protocol/scan protect ./tools.json --apply
Add EMILIA to Claude/Cursor/Cline: npx -y @emilia-protocol/mcp-server

Usage

• •Protect one privileged MCP tool call; missing evidence is refused (release_payment, delete_repo, deploy_production examples refuse without a receipt)
• •Set a finite mandate once (mission, limits, evidence, expiry) and let the agent work unattended inside it
• •Verify any receipt in-browser at emiliaprotocol.ai/verify — nothing is uploaded

Key features

• ✓Gate enforces each consequential unit of work at the executor/system-of-record boundary (Consequence Firewall): verifies required authority, refuses before mutation, consumes accepted authority once
• ✓Native evidence verification (Trust Receipt, Trust Profile, Trust Decision objects), one-time admission, and portable work records; exact-action CAID binding
• ✓Same-team cross-language verifiers (JS/Python/Go) and formal evidence (TLA+, Tamarin, Alloy); published as IETF Internet-Drafts
• ✓MCP server integration with refusal, exact-action sign, admit-once, and forged-evidence-rejected demo flows

Best for

Enterprises that must let autonomous agents execute real consequences (payments, PRs, production deploys) while keeping authority, evidence, and one-time admission deterministic and auditable.

Caveats

• ⚠Commercial Gate/Assurance Plane are separate; EMILIA is not an auditor or accredited certifier and no public certification program is operating
• ⚠Production requires a durable provenance ledger, shared atomic consumption store, pinned keys, wrapper on every path to the provider credential
• ⚠A lost response becomes INDETERMINATE — a state to reconcile, not permission to retry