The trusted conversation layer between organisations.

An agent-to-agent trust layer sits between the AI agents of two different organisations and establishes three things before any data moves: who each party is, what they are permitted to exchange, and what was actually said. Without it, an agent calling an external endpoint is trusting a URL. MCPLayer verifies the legal entity behind every agent, enforces policy inline on each message, and seals the exchange to a tamper-evident audit ledger.

Most MCP gateways solve an internal problem: routing, rate-limiting and securing the MCP servers your own developers connect to. They assume both ends belong to you.

MCPLayer solves the cross-organisation problem. It assumes the counterparty is a different legal entity - another bank, an insurer, a government agency - and supplies what an internal gateway never has to: entity verification, licence checks, per-counterparty policy, and a shared audit record both sides and their regulators can rely on.

Registration is not self-service assertion. An organisation joining the network passes legal entity checks, proves domain control, and has its regulatory licences verified. Only then is a cryptographic attestation issued and bound to its published MCP endpoints. When your agent connects, it is checking that attestation - not a hostname.

Agent attestation is cryptographic proof that a given AI agent is operated by a specific, verified organisation and is acting within a declared scope. It is the difference between an agent that claims to represent an insurer and one that provably does. Attestations are revocable instantly, so a compromised or decommissioned agent loses access across the network at once.

Every exchange is sealed to an immutable, replayable ledger: both parties' identities, the policy decision applied, the payload class, and the outcome. When an agent-negotiated result is disputed, you hand over a record rather than an explanation. "The model decided" is not evidence; a signed, timestamped transcript is.

APRA's existing standards already reach agent behaviour - CPS 230 on operational resilience and CPS 234 on information security - and APRA has specifically flagged the misuse of autonomous AI agents and insecure integrations as attack pathways.

MCPLayer supplies the control evidence those standards ask for at the organisational boundary: verified counterparty identity, enforced data-class policy, instant revocation, and a complete record of every exchange. It is designed for the vendor risk questionnaires Australian regulated entities actually send.

No. Bring the MCP servers you already run. You bind them to your organisation's attestation, declare the data classes and scopes they expose, and set counterparty policy per vertical or per organisation. The servers stay yours; MCPLayer governs who may reach them and on what terms.

Australian regions first, with per-organisation key management and rotation, and encryption in transit and at rest. Sensitive data classes can be configured for zero-retention, where the exchange is attested and recorded in the ledger without the payload itself being persisted.