noordeen123/culprit

Overview

A deterministic, offline, read-only debugging tool for AI agents. It identifies which change introduced a bug and assesses whether a proposed fix is complete. Built as an MCP-native Python tool, it integrates with agent workflows without modifying the codebase.

Best for

Best for
Developers using AI agents who need reliable, non-intrusive debugging assistance.

Use cases

• Pinpoint the commit or change that introduced a regression
• Verify a fix addresses the root cause completely
• Provide read-only debugging context to AI coding agents

How to use

Install

uvx culprit                     # run from PyPI on demand

Tools exposed

• verify_fix
• find_suspects
• check_completeness
• get_intent
• get_evolution
• get_risk_score
• get_blast_radius
• get_test_impact
• classify_change
• from_trace

Tested with

Claude Code, Cursor, Windsurf, Cline, Continue, VS Code

Notes

A deterministic, offline, read-only debugging tool for AI agents. It identifies which change introduced a bug and assesses whether a proposed fix is complete. Built as an MCP-native Python tool, it integrates with agent workflows without modifying the codebase.

6 stars on GitHub. Last updated 2026-08-26. Licensed MIT.

Use cases

• Pinpoint the commit or change that introduced a regression
• Verify a fix addresses the root cause completely
• Provide read-only debugging context to AI coding agents

Pros

• Deterministic and offline, ensuring privacy and reproducibility
• Read-only design avoids unintended side effects on the codebase
• MCP-native integration simplifies use with agent frameworks

Cons

• Limited adoption (6 stars) suggests early-stage maturity
• Requires MCP-compatible environment to leverage its features
• Scope may be narrow, focusing on bug introduction and fix completeness

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.