Guardrail MCP Server

Professional Model Context Protocol server for Guardrail AI.

"Stop shipping pretend features."

Installation

npm install -g guardrail-mcp-server

Configuration

See MCP-INSTALLATION-GUIDE.md for editor-specific setup.

Development

cd mcp-server
npm install
npm start

Premium Command Palette Tools

These tools provide a top-notch, zero-friction UX:

Ship Check Commands

• run_ship - Guardrail: Ship Check (GO/NO-GO)
• run_reality - Guardrail: Run Reality Mode
• run_mockproof - Guardrail: Run MockProof Gate
• run_airlock - Guardrail: Run Airlock (SupplyChain)

Report & Artifact Commands

• get_last_run - Guardrail: Open Last Run Report
• open_artifact - Open Report/Replay/Trace/SARIF/Badge
• rerun_last_check - Guardrail: Re-run Last Check
• export_sarif - Export findings as SARIF

Setup & Policy Commands

• run_doctor - Guardrail: Doctor (Fix my setup)
• edit_policies - Guardrail: Policies (Quick Edit)
• get_status - Get server status and workspace info
• policy_patch - Apply atomic policy changes

Fix Mode Commands

• enter_fix_mode - Enter Fix Mode for blocker resolution
• fix_mode_status - Get Fix Mode checklist status
• mark_fix_complete - Mark blocker as fixed
• exit_fix_mode - Exit and re-run ship check

Evidence & Diagnostics

• explain_finding - Get detailed finding explanation

AI Guardrail Tools (Prompt Firewall + Output Verification)

These tools provide AI safety and verification capabilities:

| Tool | Description | |------|-------------| | guardrail.verify | 🛡️ Verify AI agent output before applying - checks secrets, dangerous commands, path traversal | | guardrail.quality | 📊 Code quality analysis - complexity, maintainability, technical debt metrics | | guardrail.smells | 👃 Code smell detection - anti-patterns, naming issues, structural problems | | guardrail.hallucination | 🔍 Hallucination check - verify claims against actual source code | | guardrail.breaking | ⚠️ Breaking change detection - API changes, removed methods, type changes | | guardrail.mdc | 📝 MDC Generator - source-anchored documentation generation | | guardrail.coverage | 🧪 Test coverage mapping - identify untested components |

Example Usage

// Verify AI-generated code before applying
{
  "tool": "guardrail.verify",
  "arguments": {
    "input": "{\"format\":\"guardrail-v1\",\"diff\":\"...\",\"commands\":[]}",
    "mode": "build"
  }
}

// Check code quality
{
  "tool": "guardrail.quality",
  "arguments": {
    "projectPath": ".",
    "threshold": 70
  }
}

// Detect code smells
{
  "tool": "guardrail.smells",
  "arguments": {
    "projectPath": ".",
    "severity": "high"
  }
}

Agent Checkpoint Tools

Pre-write validation that blocks AI agents until issues are fixed:

| Tool | Description | |------|-------------| | guardrail_checkpoint | 🛡️ Validate code before writing - blocks on TODOs, mocks, console.log, etc. | | guardrail_set_strictness | ⚙️ Set checkpoint strictness: chill, standard, strict, paranoid | | guardrail_checkpoint_status | 📊 Get current checkpoint status and blocking violations |

Architect Tools

AI agents consult the Architect before writing code:

| Tool | Description | |------|-------------| | guardrail_architect_review | 🏛️ Review code against architecture patterns | | guardrail_architect_suggest | 💡 Get architectural guidance before writing code | | guardrail_architect_patterns | 📋 List all active architecture patterns | | guardrail_architect_set_strictness | ⚙️ Set architect strictness level |

Codebase Architect Tools

Deep codebase knowledge for AI agents:

| Tool | Description | |------|-------------| | guardrail_architect_context | 🧠 Load full codebase context (tech stack, conventions, patterns) | | guardrail_architect_guide | 🏛️ Get guidance for creating/modifying code | | guardrail_architect_validate | ✅ Validate code against codebase patterns | | guardrail_architect_dependencies | 🔗 Understand file relationships and impact |

Guardrail 2.0 Tools (Consolidated)

Six core tools for the complete workflow:

| Tool | Description | |------|-------------| | checkpoint | 🛡️ Block AI agents until issues are fixed (pre/post write) | | check | 🔍 Verify code is real, wired, honest | | ship | 🚀 Go/No-Go decision (GO / WARN / NO-GO) | | fix | 🔧 Fix blocking issues safely | | status | 📊 Health + version info | | set_strictness | ⚙️ Set checkpoint strictness level |

Intent Drift Guard Tools

Capture intent before writing code, monitor for drift:

| Tool | Description | |------|-------------| | guardrail_intent_start | 🎯 Start a new step with explicit intent | | guardrail_intent_check | ✅ Check if code changes align with stated intent | | guardrail_intent_validate_prompt | 🔒 Validate new prompts against locked intent | | guardrail_intent_status | 📊 Get current Intent Drift Guard status | | guardrail_intent_complete | ✅ Complete step and generate proof artifact | | guardrail_intent_lock | 🔒 Lock intent to prevent scope expansion | | guardrail_intent_unlock | 🔓 Unlock intent, allow scope changes |

Core Analysis Tools

• validate_project - Validate project structure and API endpoints
• check_design_system - Validate design system consistency
• check_project_drift - Check for architecture drift
• setup_design_system - Set up and lock design system
• register_api_endpoint - Register API endpoint
• get_project_health - Get project health score
• get_guardrails_rules - Get guardrails rules
• architect_analyze - Intelligent project analysis
• build_knowledge_base - Build codebase knowledge
• semantic_search - Search code by meaning
• security_scan - Full security scan
• ship_check - Ship readiness check
• get_deploy_verdict - Get deploy GO/NO-GO decision

Resources

• guardrails://rules - Guardrails rules document
• guardrails://templates - Available templates
• guardrails://design-tokens - Design system tokens

Documentation

See MCP-PREMIUM-TOOLS.md for detailed tool documentation.

Privacy & Trust

• Runs locally
• Artifacts saved to .Guardrail/
• No upload unless you export/share