GuardRail Security MCP
Security MCP · v2.1 · MIT open core

Static security analysis for AI-assisted development.
MCP tools agents can call before merge.

Hybrid multi-language security for Cursor, Claude, and CI. Secrets, multi-hop taint, tree-sitter sinks, repo/PR scans, SARIF/SBOM, and enterprise policy ALLOW/DENY — all as MCP tools agents can call.

View on GitHub Run interactive demo See pricing
0MCP tools
0automated tests
0languages
4hybrid engines
GuardRail security dashboard mockup with findings and risk score
CRITICAL · SQL taintrequest → execute
Secrets redactedsafe for agent chat
Cursor Claude Desktop VS Code GitHub Actions MCPize Smithery-ready SARIF · CycloneDX · SPDX Enterprise RBAC
Scope: static analysis MCP · does not execute scanned code · website playground is browser-only · see docs/ACCURACY.md

Built for the agent development loop

GuardRail sits between AI-generated patches and your main branch — same engines in the IDE and CI.

AI agent scanning code

Agent-native review

MCP tools return redacted findings + remediations models can apply. No secret leakage into chat history.

Multi-language security

Multi-language hybrid scan

Python AST + multi-hop taint, tree-sitter for JS/Go/Java/Rust/C/C++, plus high-signal language grids.

CI CD security gate

PR & CI gates

Git diff mode, SARIF upload, SBOM export, policy packs that emit ALLOW/DENY for release control.

Enterprise control plane

Enterprise control plane

API keys/JWT, RBAC, path sandbox, quotas, audit logs, compliance evidence, Prometheus metrics.

Four engines. One clean report.

Complementary analyzers reduce blind spots. Results are deduped so agents get signal, not noise.

RX

Secret & pattern grid

Keys, tokens, PEM, TLS verify-off, shell injection patterns, provider prefixes.

AST

Python semantic AST

eval aliases, shell=True, pickle, insecure YAML, SQL interpolation with line evidence.

TN

Multi-hop taint

Tracks untrusted data across assignments into SQL, command, eval, path, and SSRF sinks.

TS

Tree-sitter CST

Structural sinks in Python, JS, Go, Java, C/C++, Rust — stronger than pure text search.

PL

Plugins & rules

Org YAML/JSON rules and Python plugins without forking the core server.

CI

Repo · Diff · Cache

Parallel walks, PR-only scans, incremental content-hash cache for monorepos.

audit_code_safety scan_repository scan_git_diff scan_dependencies export_sarif generate_sbom suggest_fixes evaluate_policy full_pipeline enterprise_health

Why teams pick GuardRail

Directories list tools. GuardRail is a full security workflow for agent-written code.

Capability Generic MCP toy Cloud SAST only GuardRail
Works inside Cursor / Claude MCP ✓ Often no ✓
Multi-hop taint + tree-sitter — Sometimes ✓
Secret redaction for agent chat Rare Varies ✓
SARIF + SBOM + PR diff mode — ✓ ✓
Self-host free / MIT core Sometimes — ✓
Enterprise RBAC + policy packs — $$$$ ✓ included

How it fits your workflow

From IDE agents to production CI — one server, consistent verdicts.

Install the MCP

Point Cursor / Claude / VS Code at python -m guardrail --mode stdio, or deploy on MCPize.

Scan before apply

Agents call audit_code_safety or scan_git_diff on generated patches and pull requests.

Gate with policy

Strict / SOC2 / PCI packs return ALLOW or DENY. Export SARIF and compliance evidence for audits.

Live security playground

Paste code or load a sample. Browser-side demonstration using representative rule IDs. Not the Python runtime. Use the MCP/CLI for production scans.

Tip: edit the code and re-run. Secrets in excerpts are redacted.

Scan result

Browser demonstration

—
VerdictReady
Policy—
Issues—
Risk score—
Grade—
Enginesdemo-regex + taint
Findings0
Run a scan to see findings with line numbers, severity, and remediations.

Ground truth CLI: python -c "from guardrail.hybrid_scan import hybrid_scan; …" · Ground truth: Python hybrid_scan / MCP tools

Pricing that matches security value

Self-host free forever. Monetize on MCPize with freemium → Pro → Enterprise.

Free
$0/mo
Open source · local MCP
  • Full hybrid engines (self-host)
  • STDIO for Cursor / Claude
  • 65 tests + examples
  • MIT license
Clone free
Pro
$29/mo
Hosted Pro plan (MCPize)
  • Hosted MCP endpoint
  • Repo / PR / SARIF / SBOM
  • Fix drafts
  • Marketplace listing
Start Pro
Team
$99/mo
Higher limits · policy packs
  • Everything in Pro
  • Higher scan limits
  • Policy packs
  • Shared team access
Choose Team
Enterprise
$299/mo
RBAC · audit · sandbox
  • RBAC · JWT · audit logs
  • Path sandbox & quotas
  • Compliance evidence
  • Custom policy support
Enterprise docs

Install in minutes

Open source on GitHub. STDIO for IDEs. HTTP for dashboards and hosted gateways.

Cursor / Claude config

mcp.json 
{
  "mcpServers": {
    "guardrail": {
      "command": "python",
      "args": ["-m", "guardrail", "--mode", "stdio"],
      "cwd": "/path/to/guardrail-mcp",
      "env": { "PYTHONPATH": "/path/to/guardrail-mcp" }
    }
  }
}

Clone & verify

terminal 
git clone https://github.com/SECRET4422/guardrail-mcp.git
cd guardrail-mcp
pip install -r requirements.txt
export PYTHONPATH=$PWD
python -m unittest discover -s tests -v
python -m guardrail --mode stdio

FAQ

Straight answers for buyers and security reviewers.

Abstract security illustration
Is GuardRail a full commercial SAST?
It’s a hybrid heuristic + semantic MCP guardrail optimized for agents and CI. It is not a substitute for enterprise SAST suites, but it’s far beyond toy regex scanners.
Will secrets leak into the LLM chat?
Findings use redacted excerpts. Still: never scan production secrets you can’t rotate, and prefer local STDIO over pasting secrets into remote hosts.
Can I use it only in CI?
Yes. Use scan-repo, scan-diff, SARIF export, and the included GitHub Actions workflows without any IDE.
How do I make money with it?
Self-host free, or publish on MCPize with Pro/Team/Enterprise tiers. See docs/MCPIZE.md in the repo.
What languages are supported?
Deep Python (AST + taint). Structural tree-sitter for JS/TS, Go, Java, C/C++, Rust. Pattern grids for shell, Docker, Terraform, K8s, and more.

Repository & listing

Source, issues, and the MCPize listing are public. Stars and reviews are optional.

View on GitHub MCPize listing Quality checklist

Run GuardRail locally or via MCP

MIT-licensed core. Optional hosted listing on MCPize. Enterprise controls documented in-repo.

Open GitHub repo MCPize listing This website