Skip to main content
Glama
agent-security-scanner-mcp

agent-security-scanner-mcp

by sinewaveai

agent-security-scanner-mcp

Security scanner for AI coding agents, MCP servers, prompts, and AI-suggested packages.

Run it before Claude Code, Cursor, Windsurf, Cline, OpenCode, or another agent trusts new code, tools, prompts, or dependencies.

npm downloads npm version License: MIT CI

Copy-Paste Start

Scan any repo and get an A-F agent security grade:

npx agent-security-scanner-mcp scan-project . --verbosity compact

Install the scanner into your AI coding client:

npx agent-security-scanner-mcp init claude-code

Replace claude-code with cursor, claude-desktop, windsurf, cline, kilo-code, opencode, or cody.

Audit an MCP server before adding it to an agent:

npx agent-security-scanner-mcp scan-mcp ./path/to/mcp-server --verbosity compact

Check AI-generated imports for package hallucinations:

npx agent-security-scanner-mcp scan-packages ./src/app.ts npm --verbosity compact

Related MCP server: AgentAudit

What It Catches

Risk

Why agents need it

Command

Vulnerable generated code

Agents can introduce SQL injection, XSS, command injection, unsafe crypto, and secrets

scan-project, scan-security, scan-diff

MCP server attacks

MCP tools can poison descriptions, spoof names, exfiltrate env vars, or execute commands

scan-mcp

Prompt injection

Agents often process untrusted docs, tickets, pages, and tool output

scan-prompt

Unsafe agent actions

Catch dangerous shell/file/network actions before execution

scan-action

Hallucinated packages

AI often invents dependency names that attackers can later squat

check-package, scan-packages

SBOM and CVEs

Generate CycloneDX SBOMs and scan dependencies with OSV.dev

sbom-generate, sbom-vulnerabilities

Semantic review

LLM-powered review that uses project intent to find context-aware issues

cr-agent

Screenshots

Project Scan

Project scan demo

MCP Server Audit

MCP audit demo

Package Hallucination Detection

Package hallucination demo

Demos

Run a safe local demo that creates intentionally vulnerable fixtures, scans them, and cleans them up.

# MCP audit demo: tool poisoning, spoofed tool name, command execution, secret exposure
npx agent-security-scanner-mcp demo --type mcp --no-prompt

# Package hallucination demo: real imports mixed with fake AI-generated package names
npx agent-security-scanner-mcp demo --type packages --no-prompt

Expected demo shape:

{
  "grade": "F",
  "findings_count": 8,
  "findings": [
    {
      "rule": "mcp.description-injection",
      "severity": "ERROR",
      "message": "Tool description contains imperative language directed at the LLM."
    },
    {
      "rule": "mcp.tool-name-spoofing",
      "severity": "ERROR",
      "message": "Tool name is close to a well-known MCP tool name."
    }
  ]
}

Install In Your Agent

npx agent-security-scanner-mcp init claude-code

Client

Setup

Claude Code

npx agent-security-scanner-mcp init claude-code

Cursor

npx agent-security-scanner-mcp init cursor

Claude Desktop

npx agent-security-scanner-mcp init claude-desktop

Windsurf

npx agent-security-scanner-mcp init windsurf

Cline

npx agent-security-scanner-mcp init cline

Kilo Code

npx agent-security-scanner-mcp init kilo-code

OpenCode

npx agent-security-scanner-mcp init opencode

Cody

npx agent-security-scanner-mcp init cody

Interactive picker

npx agent-security-scanner-mcp init

Then restart the client. Your agent can call the scanner as an MCP tool.

Agent Playbook

Paste this into an agent task when you want it to work safely:

Before trusting new code, dependencies, prompts, or MCP tools, run:

1. npx agent-security-scanner-mcp scan-project . --verbosity compact
2. npx agent-security-scanner-mcp scan-packages ./src/app.ts npm --verbosity compact when imports change
3. npx agent-security-scanner-mcp scan-mcp ./path/to/mcp-server --verbosity compact before adding MCP servers
4. npx agent-security-scanner-mcp scan-diff --base main --target HEAD before opening a PR

Fix high-confidence security findings before continuing.

Common Workflows

Before You Trust An Agent-Written PR

npx agent-security-scanner-mcp scan-diff --base main --target HEAD --verbosity compact
npx agent-security-scanner-mcp scan-packages ./src/app.ts npm --verbosity compact

Before Installing An MCP Server

npx agent-security-scanner-mcp scan-mcp ./path/to/mcp-server --verbosity compact

Before Adding A Dependency Suggested By AI

npx agent-security-scanner-mcp check-package express npm
npx agent-security-scanner-mcp scan-packages ./package.json npm --verbosity compact

Add CI

npx agent-security-scanner-mcp init-ci github

Generate Share Copy From A Real Scan

npx agent-security-scanner-mcp scan-project . --verbosity compact > scan-result.json
npx agent-security-scanner-mcp share-kit --scan-result scan-result.json --output share-kit.md

CLI Reference

Command

Use

scan-project <dir>

Full project scan with A-F grade

scan-security <file>

Single-file security scan

scan-diff --base main --target HEAD

Scan changed files only

scan-mcp <path>

Audit an MCP server before install

scan-prompt "<text>"

Detect prompt injection and jailbreak attempts

scan-action <type> <value>

Pre-execution safety check for shell/file/network actions

check-package <name> <ecosystem>

Verify one package exists

scan-packages <file> <ecosystem>

Verify imports/dependencies in a file

doctor

Check local setup health

quickstart --client cursor

Generate repo-specific next steps

share-kit

Generate public-safe launch/share copy

export-vanta

Export scan evidence to the ProofLayer Vanta integration

All scanner outputs support context-friendly verbosity:

--verbosity minimal   # counts only, best for CI
--verbosity compact   # default, best for agents and humans
--verbosity full      # audit/debug detail

MCP Tools For Agents

When installed as an MCP server, agents get these tool families:

Tool family

Purpose

scan_security, fix_security, scan_git_diff, scan_project

Code and repo security

scan_mcp_server, scan_skill

MCP and AI skill security

scan_agent_prompt, scan_agent_action

Prompt/action safety

check_package, scan_packages

Package hallucination detection

sbom_generate, sbom_scan_vulnerabilities, sbom_diff, sbom_export_report

Supply-chain and release evidence

get_compliance_controls, evaluate_compliance

SOC2/GDPR/AIUC-1 technical evidence

scanner_health

Runtime diagnostics

Why Developers Install It

  • One npx command gives an A-F security grade for AI-written code.

  • MCP-specific checks catch risks that normal SAST tools miss.

  • Package hallucination detection checks 4.3M+ package names across npm, PyPI, RubyGems, crates.io, pub.dev, CPAN, and raku.land.

  • Output is compact by default so coding agents can read it without burning the context window.

  • Works as a CLI, MCP server, GitHub Action workflow, Apify Actor, and semantic review agent.

  • MIT licensed.

Semantic Code Review

cr-agent is bundled with the npm package for LLM-powered semantic review. It reads project intent, then looks for code that violates that intent.

npx cr-agent analyze ./path/to/project -p claude-cli --verbose
npx cr-agent analyze ./path/to/project -p openai --format sarif

Use it when rule-based scanning is not enough and the question is, "Does this code make sense for what this project is supposed to do?"

SBOM And Compliance Evidence

npx agent-security-scanner-mcp sbom-generate .
npx agent-security-scanner-mcp sbom-vulnerabilities .
npx agent-security-scanner-mcp sbom-check-hallucinations .
npx agent-security-scanner-mcp evaluate-compliance . --framework soc2-technical

Useful before releases, SOC 2 evidence collection, vendor reviews, and AI-generated dependency changes.

Apify Actor

Run the scanner as an Apify Actor when you want a hosted API or scheduled scans:

{
  "actorId": "folkloric_morale/agent-security-scanner",
  "input": {
    "target": "repository",
    "repoUrl": "https://github.com/your-org/your-agent.git",
    "repositoryScanMode": "quick",
    "includeTestFiles": false,
    "maxRepositoryFiles": 150,
    "severityThreshold": "medium"
  }
}

Latest Release

v4.5.8 adds diff-scoped cr-agent semantic review, fixes nested repository diff lookups, and filters semantic findings to changed hunks so PR reviews stay focused on code that actually changed.

Detailed release history lives in CHANGELOG.md. The older README-embedded changelog is archived at archive/README_CHANGELOG_ARCHIVE.md.

FAQ

Is this only for MCP servers? No. It scans normal repos, diffs, prompts, actions, packages, SBOMs, MCP servers, and AI skills.

Does it send my source code anywhere? Rule-based CLI and MCP scans are local. cr-agent uses the provider you choose for semantic review.

Is it a replacement for npm audit? No. It complements npm audit by catching AI-specific risks: hallucinated packages, prompt injection, MCP tool poisoning, unsafe agent actions, and vulnerable generated code.

What should I run first? Run npx agent-security-scanner-mcp scan-project . --verbosity compact.

License

MIT

Available Tools

22 tools
check_packageB

Check if a package name is legitimate or potentially hallucinated (AI-invented)

Inputs: ecosystem (required), package_name (required).

clawproof_healthA

Alias for scanner_health (deprecated, use scanner_health instead)

Inputs: no named parameters.

evaluate_complianceA

Evaluate a project against compliance frameworks (SOC2-technical, GDPR-technical, AIUC-1). Collects evidence from code scans, SBOM, vulnerability checks, and hallucination detection, then evaluates controls. Optionally saves timestamped evidence bundle.

Inputs: directory_path (required), sbom_path, verbosity, frameworks, +2 more.

fix_securityB

Scan a file and return fixes. Use verbosity='minimal' for summary only, 'compact' (default) for fix list, 'full' for complete fixed file content.

Inputs: file_path (required), verbosity.

get_compliance_controlsA

Look up compliance controls with evaluation criteria. Supports multiple frameworks: aiuc-1 (default), soc2-technical, gdpr-technical. Filter by domain, control IDs, or OWASP LLM tags.

Inputs: domain, framework, verbosity, control_ids, +1 more.

list_package_statsB

List statistics about loaded package lists for hallucination detection

Inputs: no named parameters.

list_security_rulesC

List all available security fix templates and their descriptions

Inputs: no named parameters.

sbom_check_hallucinationsA

Check all packages in an SBOM against official registries to detect hallucinated (AI-invented) package names. Supports npm, pypi, rubygems, dart, perl, raku, crates. Go/Java marked as unsupported.

Inputs: sbom_path, verbosity, directory_path.

sbom_diffA

Compare current project SBOM against a stored baseline. Reports added, removed, and version-changed packages. Use save_baseline=true to create initial baseline.

Inputs: directory_path (required), verbosity, baseline_path, save_baseline.

sbom_export_reportA

Generate an HTML or JSON audit report from SBOM data, optionally enriched with vulnerability scan results. Suitable for PCI-DSS and compliance audits.

Inputs: format, sbom_path, verbosity, output_path, +2 more.

sbom_generateA

Generate a CycloneDX v1.5 SBOM for a project. Discovers all dependencies (direct + transitive) from lock files and manifests across Node.js, Python, Go, Rust, Ruby, Java. Use verbosity='minimal' for counts, 'compact' (default) for component list, 'full' for complete CycloneDX JSON.

Inputs: directory_path (required), verbosity, include_dev, output_path.

sbom_scan_vulnerabilitiesA

Cross-reference SBOM components against OSV.dev vulnerability database. Returns CVE IDs, CVSS scores, severity, and fix recommendations. Accepts directory_path (generates fresh) or sbom_path (loads saved artifact).

Inputs: sbom_path, verbosity, directory_path, severity_threshold.

scan_agent_actionA

Pre-execution security check for agent actions (bash, file_write, file_read, http_request, file_delete, cron, process_spawn, git, docker). Returns ALLOW/WARN/BLOCK.

Inputs: action_type (required), action_value (required), verbosity.

scan_agent_promptB

Scan a prompt for malicious intent. Returns BLOCK/WARN/LOG/ALLOW. Use verbosity='minimal' for action only, 'compact' (default) for findings, 'full' for audit details.

Inputs: prompt_text (required), context, verbosity.

scan_git_diffA

Scan git diff for new security vulnerabilities. Only reports issues on changed lines. Use for PR reviews.

Inputs: base_ref, verbosity, target_ref.

scan_mcp_serverA

Scan an MCP server's source code for security vulnerabilities: overly broad permissions, missing input validation, data exfiltration, insecure patterns. Returns grade (A-F) and recommendations.

Inputs: server_path (required), manifest, verbosity, update_baseline.

scanner_healthB

Check plugin health: engine status, daemon status, package data availability

Inputs: no named parameters.

scan_packagesA

Scan code for package imports and check for hallucinated (AI-invented) packages. Use verbosity='minimal' for counts, 'compact' (default) for flagged packages, 'full' for all details.

Inputs: ecosystem (required), file_path (required), verbosity.

scan_projectA

Scan an entire directory for security vulnerabilities with .gitignore support and security grading. Use verbosity='minimal' for grade + counts, 'compact' (default) for top issues, 'full' for all details.

Inputs: directory_path (required), diff_only, recursive, verbosity, +3 more.

scan_securityC

Scan a file for security vulnerabilities. Use verbosity='minimal' for counts only (~50 tokens), 'compact' (default) for actionable info (~200 tokens), 'full' for complete metadata.

Inputs: file_path (required), engine, verbosity, output_format, +3 more.

scan_skillA

Deep security scan of an OpenClaw skill. Multi-layer analysis: prompt injection detection, code analysis (AST+taint), ClawHavoc malware signatures, package supply chain verification, rug pull detection. Returns security grade A-F with detailed findings.

Inputs: skill_path (required), baseline, verbosity.

score_aivssA

Score findings using OWASP AIVSS v2. Accepts any scanner output or raw findings JSON. Returns per-finding AIVSS scores (0-10) and aggregate posture. Use verbosity='minimal' for posture only, 'compact' (default) for scores, 'full' for all metrics.

Inputs: findings (required), overrides, verbosity, source_tool.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.5/5.0

    Scored across 22 tools

    Disambiguation4/5

    Tools are mostly distinct with specific targets (e.g., packages, SBOM, files, projects, skills, prompts, actions). Some overlap exists between check_package, sbom_check_hallucinations, and scan_packages, but descriptions clarify levels. The deprecated alias clawproof_health introduces minor confusion.

    Naming Consistency3/5

    Naming conventions are mixed: some use 'check_', 'evaluate_', 'fix_', 'get_', 'list_', 'sbom_', 'scan_', 'score_'. Within groups (sbom_*, scan_*) it is consistent, but overall there is no single verb_noun pattern across all tools.

    Tool Count5/5

    22 tools is well-scoped for a security scanner MCP server covering many scanning types (files, projects, skills, prompts, actions, SBOM, compliance). Each tool serves a distinct purpose without being excessive.

    Completeness4/5

    Covers a broad range of security scanning needs: vulnerability scanning, hallucination detection, SBOM generation and analysis, compliance evaluation, and AIVSS scoring. Minor gaps like container scanning or dependency updates, but core workflows are well-covered.

    Maintenance

    ActivityStale
    ResponsivenessLimited data

    Related MCP Connectors

    Related MCP Servers