RCPT Protocol is an open cryptographic standard for verifiable agent action receipts — signed, timestamped, tamper-evident records that any agent can generate and any system can verify. Built on W3C DIDs, Ed25519 signatures, and JSON Canonicalization Scheme (JCS, RFC 8785).

Agent did:key:z6Mk… Action tool_call Receipt ed25519:… Chain n receipts Trust 0–100
Fig. 1 — Receipt lifecycle: agent action → signed receipt → chain → trust score

Agents act without accountability. Three questions define the trust gap.

When a human sends an email, signs a contract, or executes a trade, there is an implicit accountability chain. Identity is established. The action is logged. Consequences follow. When an AI agent does the same things, there is nothing — no verifiable identity, no tamper-evident log, no standard by which one system can prove to another what happened.

Identity Who is the agent, and on whose authority does it act?
Audit What did it do, and can the record be independently verified?
Reputation Should it be trusted to do it again, based on verifiable history?

No existing standard addresses all three cohesively. RCPT Protocol is designed to close this gap.

The receipt

The atomic unit of the protocol — a cryptographically signed, tamper-evident record of a single agent action. Signed with Ed25519 (RFC 8032) over JCS-canonicalized bytes (RFC 8785). No pre-hashing. Inputs and outputs are stored as SHA-256 hashes only — RCPT is audit infrastructure, not surveillance infrastructure.

{ "rcpt_version": "0.1", "receipt_id": "01JQFK8X3YZ4A5B6C7D8E9F0", "timestamp": "2026-03-18T14:22:01.342Z", "agent_id": "did:key:z6Mkf5rGMoatrSj1f…QLP1RGvMwb5bPCG", "action_type": "tool_call", "input_hash": "sha256:a3f2b8c1d4e5f6a7b8c9d0e1f2a3b4c5…", "output_hash": "sha256:9f86d081884c7d659a2feaa0c55ad015…", "chain": { "parent_receipt_id": "01JQFK7R1…", "sequence": 3 }, "delegation": { "delegator_id": "did:web:acme.com:users:jane" }, "tool": { "name": "crm_query", "server_uri": "https://crm.acme.com/mcp/v1" }, "signature": "ed25519:3K4lRtBXkFqZ8mN2pY7vW9xA1cD5eH…" }

Identity is expressed via W3C DIDs: did:key for agents (public key is the identifier), did:web for enterprise environments with organizational anchoring. Receipt IDs are ULIDs — lexicographically sortable and globally unique. Full schema in the specification.

Proof of existence on-chain. Receipt content off-chain.

RCPT separates proof of existence from receipt content. Full receipts are stored off-chain. Merkle roots covering batches of receipts are periodically anchored to a blockchain, providing trustless timestamp proof at minimal cost. The reference implementation anchors to Solana (~$0.26 per batch of up to 10,000 receipts).

Layer 1 Identity (W3C DIDs) did:key did:web method-agnostic Layer 2 Receipt Creation JCS (RFC 8785) → Ed25519 (RFC 8032) → ULID Layer 3 Chaining & Delegation parent refs · workflow IDs · scope attenuation · DAGs Layer 4 Merkle anchoring · Solana reference · ledger-agnostic Layer 5 Trust Scoring volume · consistency · age · verification rate · disputes
Fig. 2 — Protocol layers. Layer 4 is optional; the protocol is ledger-agnostic.

Not built in isolation

RCPT complements existing identity, authorization, and transparency frameworks. It occupies the gap between IETF AIMS Layer 6 (Authorization) and Layer 7 (Monitoring) — providing cryptographic proof that an authorized action was actually performed.

W3C DIDs & VCs v2.0 IETF AIMS (draft-klrc-00) WIMSE (Workload Identity) NIST NCCoE AI Identity SPIFFE / SPIRE IETF SCITT OpenTelemetry EU AI Act (Art. 12, 19, 26)

Detailed comparison in the whitepaper §7.

Known implementations

RCPT Protocol SDKs Python · TypeScript — Reference (Apache 2.0)
MCP Server Integration MCP v1.x — Native receipt generation
Framework Integrations Anthropic · OpenAI · CrewAI · LangChain · FastAPI
Conduid Go / Solana — Production
Building an implementation? Submit a PR to be listed.

Receipt Chain Protocol for Trust

A Cryptographic Trust Standard for the Agentic Internet

This document specifies the RCPT Protocol data model, cryptographic requirements, receipt chaining semantics, agent identity model, storage architecture, delegation framework, and implementation guidance for developers and enterprises.

// BibTeX @techreport{rcpt2026, title = {RCPT: Receipt Chain Protocol for Trust}, author = {RCPT Labs}, year = {2026}, version = {0.6}, url = {https://rcptprotocol.com/whitepaper/rcpt-protocol-whitepaper-v0_6.pdf} }

RCPT Protocol v0.x is governed by RCPT Labs. Upon v1.0 stability, stewardship will transfer to an appropriate open standards body. The protocol uses semantic versioning: minor increments are backwards compatible (new optional fields); major increments may introduce breaking changes with a minimum six-month migration window. Proposed changes follow an RFC process with 30-day public comment periods via GitHub.