Research · Architecture

AxoneOS as the governance layer for MCP agent interoperability

MCP standardizes the wire format. It does not standardize authorship, authority, accountability, or dispute. A governance layer must be the part that outlives any single transport — the institution under the protocol, not the protocol itself.

Technologies compete. Institutions endure.

The Model Context Protocol is a powerful, rapidly-adopted transport — and a transport is exactly what its name promises. What MCP gives us is a common grammar for tool calls between agents and resources. What MCP does not give us is an answer to who is allowed to do what, under which rules, with which consequences, and how disputes are settled when it goes wrong.

AxoneOS treats MCP as transport. AxoneOS adds the institution: Zones as bounded jurisdictional rulesets, Pactum as opposable on-chain settlement, and an IBC-anchored auditable trail that records every act and its effects. For the full developer-facing surface, see docs.axone.xyz.

Why an institution under the protocol

MCP moves bytes and arguments. AxoneOS decides whether those moves are authorized, records them on chain, and gives every counterparty a standing rule and a route of recourse. Below: the architectural primitives the layer sits on.

Architecture

Transport vs. institution

Zones are jurisdiction-shaped Prolog regimes — the rule and the boundary — operated by identifiable operators and replicated over IBC. An act addressed to a Zone is decidable by the regime's rules, with effects that settle on chain.

Pactum is the Solidity contract family that makes those effects opposable: a counterparty can always point at a published rule, an evidence trail, and a settled outcome. Disputes are not a chat — they are a verification.

The IBC-anchored auditable trail is the part that outlives any single transport. Replace MCP tomorrow and the trail still answers who did what, under which Zone, with what evidence.

MCP is treated as transport: an excellent, current one. The architecture assumes the wire changes — the institution does not. For the developer surface and reference SDKs, see docs.axone.xyz.

The case, in two long-form pieces

Two articles — one in English, one in French — make the argument end-to-end. Both land on the same conclusion: MCP is necessary, not sufficient.

The Real Missing Layer: Why MCP Is Not Enough for Multi-Agent Systems

MCP lets agents talk to tools. Axone lets agents conclude agreements, exchange value, and cooperate under enforceable rules.

Read the EN article →

La couche vraiment manquante : pourquoi MCP ne suffit pas pour les systèmes multi-agents

MCP permet aux agents d'appeler un outil. Il ne leur permet pas de conclure un accord, d'échanger de la valeur, ni de coopérer sous des règles opposables. C'est la couche manquante, et c'est ce qu'Axone ajoute.

Lire l'article FR →

Read the research

The blog is the public research surface for governance and architecture on AxoneOS. Continue with both long-form pieces above — or browse the wider blog.