ARTFEED — Contemporary Art Intelligence

Conleash: Lattice-Based Authorization for MCP Tool Invocations

ai-technology · 2026-05-13

Conleash, a novel middleware system, tackles consent fatigue within the Model Context Protocol (MCP) by implementing boundary-scoped authorization. It incorporates a risk lattice for automatic approval of safe calls, a policy engine that allows users to define invariants, and a refinement loop that transforms user choices into reusable rules. Tested against 984 real-world traces, Conleash demonstrated an impressive 98.2% accuracy, successfully identified 99.4% of escalations, and introduced a mere 8.2 ms overhead. In a user study involving 16 participants, Conleash's scoped permissions were favored over conventional approaches, with users noting increased trust and fewer prompts.

Key facts

  • Conleash is a client-side middleware for MCP authorization.
  • It uses a risk lattice to auto-permit safe calls within known boundaries.
  • A policy engine enforces user-defined invariants.
  • A refinement loop converts user decisions into reusable rules.
  • Evaluated on 984 real-world traces with 98.2% accuracy.
  • Caught 99.4% of escalations.
  • Added only 8.2 ms of overhead for policy verification.
  • User study (N=16) showed preference for scoped permissions over traditional methods.

Entities

Sources