claude-critical-rules-mcp

Purpose

Claude Critical Rules MCP is an MCP (Model Context Protocol) server designed to let Claude AI automatically obtain and enforce key best practices in every conversation. It is based on an exhaustive analysis of more than 96 real failure cases from over 6 months of high-intensity production use, distilling the lessons learned into 21 numbered rules to help prevent common AI assistant mistakes.

Main Capabilities

• 21 critical rules: Covering 5 categories—verification, backup, execution, database, and advanced—forming a complete checklist to prevent common AI assistant failures.
• Pre-check verification: Provides a mandatory checklist before executing technical tasks.
• Automatic update system: Can automatically check for and install rule updates from GitHub.
• Version tracking: Supports SHA256 checksums and changelog integration.
• Evidence-based: Every rule is derived from real production failure cases.

Available Tools

• verify_compliance: Displays the mandatory 21-rule checklist before technical tasks, including a 21-point checklist organized by category, rule numbers and descriptions, confirmation requirements, and version information.
• get_rules_summary: Quickly references all 21 rules and descriptions, including 5 core principles, all 21 rules listed by category, a 6-step mandatory workflow, and update instructions.
• get_version_info: Displays the current version and checks for updates, including current version details, remote version comparison, update availability status, and SHA256 hash information.
• check_for_updates: Checks whether there is a new rules version on GitHub, including version comparison and changelog references.
• Download and install the latest rules from GitHub: Includes SHA256 verification and backup location.

Available Resources

• critical-rules://instructions: The complete CRITICAL-RULES.md document, including all 21 rules, examples, and detailed explanations.
• critical-rules://changelog: The complete changelog, including version history and update details.

The 21 Rules (Quick Reference)

Verification (Rules 0-5):

• Never act before fully reading the instructions
• Search for current best practices (mandatory use of web_search)
• Read the skill before creating documentation
• Before modifyingRead the entire file
• Verify structure/location, do not assume
• Check the correct file/server

Backup and permissions (rules 6-8):

• Search prior context if mentioned
• Use timestamped backups in a standardized directory
• Ask before deleting/modifying critical items

Execution and verification (rules 9-12):

• Ask about scope before implementing
• Stop on failure (do not cascade)
• Verify with evidence
• Full logs (not just the last 20 lines)

Database and security (rules 13-15):

• Database: backup → test → verify rollback
• Distinguish production from development
• Security verification (IP, fail2ban, firewall)

Advanced rules (rules 16-21):

• Long commands (>30 seconds) → run in background
• Never use sed (use python/awk/perl instead)
• Verify file line counts before and after editing
• Follow instructions exactly
• Check MCP, skills, context first
• Mandatory 21-point confirmation

Core principles

• Verify, do not assume: always check before acting
• Back up everything: back up before any modification
• Ask permission: ask before deletions and critical changes
• Evidence-based: test with proof, not assumptions
• Stop on error: do not continue after a failure

6-step mandatory workflow

1. Ask about scope before starting
2. Full analysis
3. Plan and explain
4. Execute carefully
5. Verify with evidence
6. Document changes

How to use

Install via npx (recommended): add the configuration to claude_desktop_config.json. macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json It can also be installed globally and then configured. After configuration is complete, restart Claude Desktop to load the MCP server.

There are two mandatory steps before any technical task; see the repository documentation for example workflows.

Evidence base

A comprehensive analysis based on the following: more than 96 documented failures in production environments, more than 20 recurring patterns identified and classified, more than 6 months of intensive Claude AI usage, multiple production systems (servers, databases, APIs, deployments).

Auto-update security

• SHA256 verification: all downloaded rules undergo integrity verification
• Automatic backup: save the previous version before updating
• Version tracking: complete changelog and version history
• Manual override: force parameter for explicit reinstallation

Applicable scenarios

Suitable for users who use Claude AI for technical tasks in production environments, want to reduce common mistakes, and need mandatory verification and backup processes.

Limitations

The materials do not provide information about fees, pricing, the scope of platform compatibility, specific performance guarantees, etc. The rule content is subject to the repository documentation.

Contribution and license

Contributions are welcome: Fork the repository, create a feature branch, document failure patterns with examples, and submit a pull request. MIT License, see the LICENSE file for details.

Version: 1.2.0; Last updated: 2026-01-10; Number of rules: 21 numbered rules; Failure patterns analyzed: 96+.