ProofFlow turns AI coding sessions into auditable ledgers. It exposes MCP tools to create work contracts, record algorithm decisions, capture git snapshots, bind evidence to agent claims, and export proof packets before merging. You'd reach for this when you want reviewable trails for autonomous agent changes instead of blind trust. The server handles the full chain from contract declaration through cost budgets, diff snapshots, evidence collection, and deterministic evaluation of done criteria. It includes AgentGuard for PR reviews and can gate risky actions before execution. Install via PyPI's proofflow-mcp package and connect over stdio. The project dogfoods itself with CI workflows that post stable review comments and upload proof packets as artifacts.
claude mcp add --transport stdio io.github.hyperion-gpu-proofflow uvx proofflow