Back to Ecosystem Pulse
ToolExperimentalMCPA2A
bernstein
by sipyourdrink-ltd
Deterministic orchestrator for coding agents with worktree isolation and HMAC audit logs
Python
Updated Jul 26, 2026
Share:
How It Works
Orchestrates deterministic runs across 30+ CLI coding agents with git worktree isolation to keep agent state reproducible [Supervisor Pattern]. Creates an HMAC-signed audit trail and can run as an MCP server to expose agent-to-agent messaging Agent-to-Agent Protocol (A2A) and context sharing Model Context Protocol (MCP). Distinguishing features include deterministic orchestration, per-run sandboxing via worktrees, and cryptographic tamper-evident logs.
The Value Proposition
As agents interact and delegate, reproducibility and auditable history become essential for trust and diagnosis. Bernstein provides the primitives — isolation, deterministic execution, and an HMAC audit trail — that make comparing agent behavior and building a track record possible. That matters for multi-agent trust and A2A evaluation because you can reproduce failures, attribute actions, and feed reliable traces into reputation or benchmark systems [Guardrails Pattern].
When to Use
Teams evaluating or stress-testing multi-agent coding workflows who need reproducible runs, tamper-evident logs, and an MCP-compatible server mode.
Real-World Examples
- Isolating agent runs and reproducing multi-agent failures using git worktrees
- Creating cryptographically-signed audit trails for agent interactions and results
- Running an MCP-compatible server to record and replay agent-to-agent message flows
- Comparing agent behavior deterministically across versions or configurations
Works With
openaianthropichuggingface
Topics
a2aagent-orchestrationagentic-aiagentic-engineeringagentic-workflowai-agentsai-orchestrationai-orchestratorautonomous-agentsclaude-code+10 more
Similar Tools
autogencrewai
Keywords
multi-agent trustagent-to-agent evaluationmcpdeterministic-orchestration