Synoptis: Mutual Attestation for Trustworthy MCP Agent Networks

Read full text Summarise with VIRAT
Authors: Masaomi Hatakeyama
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The Model Context Protocol (MCP) enables AI agents to dynamically discover and invoke tools provided by external servers, creating flexible multi-agent networks. However, this dynamic capability acquisition introduces a fundamental trust problem: when agents can modify their available capabilities at runtime, static permission lists and single-protocol attestation are insufficient to ensure trustworthy interactions. Recent work on AttestMCP has shown that adding capability attestation within MCP reduces attack success rates from 52.8% to 12.4%, but relies on unidirectional, centralized verification. We present Synoptis, a mutual attestation protocol that enables MCP agents to make bidirectional, cryptographically signed claims about each other's behavior, verified against a hash-chained tamper-evident log. Synoptis introduces three contributions: (1) a bidirectional attestation protocol with a challenge-response mechanism for contestation and local trust updating, (2) a five-dimensional dynamic trust scoring model that captures quality, freshness, diversity, velocity, and revocation status, and (3) an append-only hash-chained registry that provides local, blockchain-grade integrity guarantees without requiring consensus. We implement Synoptis as a modular SkillSet within KairosChain, a self-referential MCP server framework. Through a survey of related work spanning MCP security, decentralized identity, blockchain-based trust, and multi-agent reputation, we identify that, to our knowledge, no existing system combines mutual cryptographic attestation with tamper-evident logging for MCP agent networks. We present the protocol specification, analyze its security properties against four attack scenarios (tool poisoning, Sybil attacks, collusion, and score manipulation), and discuss design trade-offs including the deliberate choice of hash-chained logs over full blockchain consensus.

Fields of Study & Subjects

Access Control and Trust Blockchain Technology Applications and Security Security and Verification in Computing

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN