CORTEXA
← Browse
arxivcs.CRcs.AI2026-06-27

From Tool Connection to Execution Control: Benchmarking Security Invariants in MCP-Style Agent Runtimes

Ting Liu

Model Context Protocol (MCP)-style ecosystems give language-model applications a practical connection layer for tools, resources, prompts, and transports. As agents move from connection to execution, security decisions often remain split across clients, servers, prompts, approval dialogs, OAuth deployments, and logs. This paper asks whether a runtime can make execution-layer invariants explicit and testable while preserving MCP-like workflows. We define eight invariants: metadata non-authority, grant-backed approval, canonical resources, principal binding, scoped capability invocation, source-and-target data-flow authorization, deny-path audit, and explicit protocol state. We implement these invariants in HCP, a Handle-Capability Protocol reference runtime for MCP-style agent execution that represents calls through principals, resources, grants, capabilities, handles, policy decisions, data-pipe checks, and audit entries. We evaluate HCP against two MCP-like baselines: a naive connection-layer runtime and a practice-informed connection-layer mitigation baseline with metadata linting, session checks, and per-call approvals. Across 10 benchmark cases, the naive baseline permits all modeled attacks, the mitigation baseline permits 6 of 10, and HCP blocks all 10 while preserving audit evidence. Ablations identify which runtime components block attacks and preserve forensic evidence. A local in-memory microbenchmark reports sub-millisecond mean latencies for measured policy, invocation, peek, and pipe operations. A bounded GitHub README-screening sample provides ecosystem signals, not vulnerability findings. The results support a narrow claim: MCP-style agent systems need an execution-control layer in addition to connection-layer conventions.

View free PDFSource page

Related papers

arxivcs.CRcs.AI2026-07-07

The Balkanization of Execution-Security Research for AI Coding Agents: Isolation, Access Control, and Time-of-Check-to-Time-of-Use Vulnerabilities

Mohammadreza Rashidi

AI coding agents now read repositories, call tools, and execute shell commands with limited human oversight, and a fast-growing body of work studies whether the execution layer around them is actually safe. That literature is scattered. Papers on sandbox isolation, capability and…

View free PDFSource page
arxivcs.CRcs.AI2026-07-03

Securing Multi-Tool AI Agent Chains With Dynamic, Real-Time Compositional Policies

Chris Schneider, Kriti Faujdar, Philipp Schoenegger, Ben Bariach

Modern AI agent implementations such as frontier coding agents chain multiple tools at runtime that create a security surface that per-tool guardrails are unable to address, as individually permitted tools can violate organizational policies when composed. We propose the Dynamic…

View free PDFSource page
arxivcs.GTcs.AIcs.CRcs.LGcs.MA2026-07-10

A Knowledge-Based Multi-Agent Framework for Security Control Recommendation

Carolina Fernández-Martínez, Shuaib Siddiqui, Vanesa Daza

Hardening IT on-premises environments can be a daunting task for teams without access to adequate cybersecurity expertise. In this regard, Decision Support Systems (DSS) with embedded expert knowledge can assist users by guiding them with security recommendations to meet their ob…

View free PDFSource page
arxivcs.CRcs.AIcs.LG2026-07-20

Adaptive Adversaries: A Multi-Turn, Multi-LLM Benchmark for LLM Agent Security

Devina Jain, David Hartmann, Chuan Li

LLM-based agents process external content, exposing them to prompt injection and multi-turn manipulation. Most safety benchmarks evaluate defenders against fixed attack pools collected before evaluation, single-turn or multi-turn. We present a 21-scenario benchmark for \emph{adap…

View free PDFSource page