v3.0.3 — Call-Edge Intelligence & Agent Adapters

AI code review CLI
that catches
review blockers

Catch review blockers, architectural violations, and security vulnerabilities before they ever reach a human reviewer. MP Sentinel is an AI Code Review CLI — multi-provider, ESM-native, runs 100% locally.

Architecture Secrets CI Gate

View on GitHub

mp-sentinel review — main...HEAD

$ npx mp-sentinel

─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─

gemini-2.5-flash MCP: github

BLOCKER src/api/users.ts:42-58

Database query inside API route handler bypasses service layer. Use UserService instead.

CRITICAL src/hooks/useData.ts:15-22

useEffect used for data fetching. Consider using a data loader or React Query pattern.

MAJOR src/lib/controller.ts:88-105

Controller contains business logic (SOLID violation). Extract to src/lib/service.ts.

Files reviewed: 24 Issues found: 10 Failed (exit code 1)

Local-First

Your code never leaves your machine unencrypted.

Multi-Provider

Gemini, GPT-5, Claude, Grok, OpenRouter — you choose.

CI-Ready

Exit-code gating for GitHub Actions, GitLab CI, and more.

Capabilities

Everything you need for better reviews

MP Sentinel goes beyond surface-level linting to catch what matters — architecture, security, and design patterns.

Multi-Provider AI

Run reviews through Gemini, GPT-5, Claude, Grok or OpenRouter. Switch providers with a single env var — no code changes needed.

Architectural Guardrails

Detect architectural violations — Controllers calling DB directly, useEffect data fetching, SOLID principle breaches — not just trivial lint errors.

3-Layer Security

Smart filtering + secret scrubbing + dry-run mode. Your secrets NEVER leave your machine unencrypted. Designed for enterprise compliance.

Repository Intelligence

Source-index aware. Understands blast radius of hub files, public API risk, test coverage gaps, and key dependency chains.

CI/CD Native

Built-in GitHub Actions + GitLab CI support. Standard exit codes (0/1/2) to gate PRs and MRs. Works in any pipeline.

Agent Adapters & Init (v3.0.3)

Guided init command for fast setup. 7+ agent adapters — Claude Code, Cursor, Copilot, Aider, Roo, Zed, JetBrains Junie — plus per-rule opt-out and expanded MCP preset library.

Pipeline

How It Works

A three-step local review loop. Install once, configure your rules, and let AI catch problems before they ship.

terminal BASH

npm install -D mp-sentinel

Install

Add MP Sentinel as a dev dependency — one command, zero config.

Works with npm, pnpm, yarn, and bun.

.mp-sentinelrc.json JSON

{
  "techStack": ["typescript", "react", "node"],
  "rules": ["no-direct-db-in-controllers"]
}

Configure

Create a config file with your tech stack, review rules, and AI provider.

Built-in rules cover architecture, security, and style.

terminal BASH

mp-sentinel
# or: mp-sentinel --format json > review.json

Review

Run mp-sentinel — defaults to diffing origin/main…HEAD. Output as console, JSON, or Markdown.

Exit code 1 = issues found. Gate PRs in CI.

Usage

CLI that fits your workflow

Review staged changes, specific commits, branches, or arbitrary files — whatever your workflow demands.

mp-sentinel — bash

mp-sentinel --staged

Review only staged changes — perfect for pre-commit hooks.

Configuration

Configure once, review everywhere

A single .mp-sentinelrc.json defines your tech stack, rules, AI provider, and MCP context.

.mp-sentinelrc.json

JSON

{
  "techStack": ["typescript", "react", "node", "postgres"],
  "rules": ["no-direct-db-in-controllers", "enforce-service-layer"],
  "ruleFiles": ["./rules/custom-rules.json"],
  "ai": {
    "provider": "google",
    "model": "gemini-2.5-flash",
    "temperature": 0.3
  },
  "mcp": {
    "servers": {
      "github": {
        "command": "npx",
        "args": ["@modelcontextprotocol/server-github"]
      }
    }
  }
}

techStack
Defines your project technologies for context-aware reviews.

rules / ruleFiles
Built-in rules plus custom paths for organization-specific guardrails.

ai provider
Switch between Google, OpenAI, Anthropic, xAI, or OpenRouter.

MCP servers
Inject external context (GitHub, docs) into review prompts.

Providers

Choose your AI provider

MP Sentinel supports multiple AI providers. Switch between them with a single environment variable — no code changes.

Balanced

Google Gemini

Best speed-to-cost ratio for most teams.

Model gemini-2.5-flash

Premium

OpenAI GPT

Highest review quality. Industry benchmark.

Model gpt-5.2

Premium

Anthropic Claude

Strong reasoning and long-context reviews.

Model claude-sonnet-4-6

Balanced

xAI Grok

Fast reasoning model. Great for CI pipelines.

Model grok-4-1-fast-reasoning

Budget

OpenRouter

Try any provider with one API key.

Model openai/gpt-5.2

Models listed above are defaults. You can override them per-provider in your .env config — e.g. "AI_MODEL"="claude-opus-4-7".

Security First

Your code stays local
— unless you say so.

A three-layer defense model: intelligent filtering, automated secret scrubbing, and dry-run previews give you full control over what reaches any AI provider.

40+ secret patterns detected

0 bytes sent without approval

3 security layers

File Filtering

Allowlist + .gitignore rules + sensitive blocklist. Only relevant files ever reach analysis.

Allowlist.gitignoreBlocklist

Secret Redaction

Detects and redacts API keys, tokens, and credentials automatically. Patterns match 40+ secret types.

API keysTokensCredentials

Dry-Run Preview

Preview token estimation and file selection before anything is sent. Zero data egress without approval.

--dry-runToken estimate

AI Provider

Only filtered, redacted, approved data is transmitted over HTTPS.

Encrypted Redacted Approved

Editors

Works with your editor

Generate agent rules for your editor of choice. MP Sentinel integrates with the tools you already use.

Cursor

Cline

Codex

Windsurf

Antigravity

FAQ

Frequently asked
questions

Quick answers to common concerns about using MP Sentinel in your workflow.

Privacy

Is my code sent to external servers?

Only the files you explicitly review are processed, and only if you configure an AI provider. MP Sentinel never sends your code anywhere without your knowledge. Use --dry-run to preview what would be sent — including token estimates — before any data leaves your machine.

AI Models

Which AI provider should I choose?

It depends on your budget and quality requirements. Google Gemini (gemini-2.5-flash) offers the best balance of speed and cost. For the highest review quality, use OpenAI GPT-5 or Anthropic Claude. OpenRouter lets you try multiple providers without changing your configuration.

CI/CD

Can I use MP Sentinel in CI/CD pipelines?

Yes. MP Sentinel ships with built-in support for GitHub Actions and GitLab CI. It uses standard exit codes: 0 for pass, 1 for issues found, 2 for errors. This makes it trivial to gate PRs and MRs in any pipeline.

Workflow

Does MP Sentinel replace human code review?

No. MP Sentinel augments human review by catching issues before the PR is opened — architectural violations, security vulnerabilities, anti-patterns — so reviewers can focus on design, correctness, and business logic rather than surface-level problems.

vs. Linting

How is MP Sentinel different from ESLint or SAST tools?

ESLint, Prettier, and traditional SAST tools operate on static syntax rules. MP Sentinel uses AI to understand intent and architecture — it catches violations that no static rule can express, like a Controller calling the database directly, or a React useEffect used for data fetching instead of a proper loader.

Custom Rules

Can I write custom review rules?

Yes. Define custom rules in .mp-sentinelrc.json under "rules" or point to external rule files with "ruleFiles". Rules can target specific patterns, architectural layers, or naming conventions unique to your project.

Open Source · Free Forever

Ship cleaner
code today.

MP Sentinel is a free, MIT-licensed AI code review CLI that catches anti-patterns, security issues, and architectural violations — before they reach production. Runs entirely on your machine. No data leaves without your approval.