DeepYardDeepYard
A

AgentCheck

Debug LLM agents with reproducible recordings and controlled interventions over MCP

Open SourceFree

About

AgentCheck is a research workbench implementing the Reproduce-Intervene-Mitigate methodology for debugging LLM agents. It records agent-tool interactions over MCP, enables replay with controlled interventions to isolate issues, and tests fixes in a sandbox before production deployment. Designed for researchers and developers building reliable agent systems who need systematic debugging workflows beyond traditional logging.

Details

Type
Integrations
Language

Tags

mcpobservabilitytool-useopen-sourceevaluationresearchdebugging