Preprint
Jul 2026
AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP
AgentCheck is presented, an open-source web workbench that turns an MCP server into an intervention surface that makes tool-using LLM failure modes reproducible, comparable, and verifiable before deployment.
Aritra Mazumder, N. J. Lia
· 2 citations
· ⚡1