M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
A

Agent Security Scanner MCP Server

Community
Security scanning for AI coding agents.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #272 of 438
Stars
★ 124 Very popular
Transport
stdio (local process)
Runtime
Node.js
Credentials
No credential needed
License
MIT
Last commit
Tools
17
49FMRS · D

An MIT-licensed open-source security-scanning MCP server focused on AI-agent-specific risks, including prompt injection, MCP tool poisoning, package hallucinations, unsafe agent actions, and vulnerable generated code. The source confirms stdio transport and initialization commands for multiple clients, but does not provide a standalone authentication configuration example.

Strongest · Documentation 12/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
10/20
Maintenance
8/20
Documentation
12/20
Setup experience
11/20
Why each score
Reliability 8/20
The README describes CLI commands, MCP tool families, and multiple scan workflows, but the supplied material contains no implementation source, tests, or CI workflows verifying MCP initialization, tool-list handshakes, error handling, or declared behavior. The manifest reports version 4.4.9 while the README identifies 4.5.8 as the latest release, creating a version contradiction. The happy path is plausible, but reproducibility of key paths is unverified.
Security and permissions 10/20
The project is explicitly positioned as a security scanner, and the README states that rule-based scans run locally while cr-agent sends code to a provider selected by the user; it also describes hosted Apify scans and Vanta export, so the data boundary is not entirely local. The supplied material does not show a permission model, credential handling, network scope, confirmation requirements for dangerous actions, or the implementation of fix_security. Points are deducted for unverified least-privilege and side-effect controls. No explicit red-line risk is established by the supplied evidence.
Maintenance 8/20
The repository is marked as not archived and provides an MIT license, version information, npm links, GitHub issues, and a changelog link, indicating basic maintenance entry points. However, no commit history, release cadence, dependency-update process, vulnerability-response channel, or maintainer accountability is provided. Stars and issue counts are discovery signals only and do not establish maintenance quality. The evidence does not support a sustained, mature maintenance score.
Documentation 12/20
The README covers installation, major commands, tool families, client integrations, example workflows, verbosity modes, data-transmission disclosure, and some limitations, making it reasonably usable. Deductions apply because exact tool parameters, authentication requirements, runtime and platform constraints, cost information, detailed troubleshooting, and executable test evidence are absent. The manifest and README also contradict each other on the version.
Setup experience 11/20
The README provides an npx installation path, init commands for several mainstream clients, and a restart instruction, so the basic setup path is clear. The supplied material does not show generated client configuration, required Node.js or platform versions, handshake verification, authentication configuration, or recovery steps for failed setup. The process may work, but important environment and manual assumptions remain unverified.

Static review · not runListed 2026-08-14

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesRuns commands or codeUses the network

Best for

  • Developers using Claude Code, Cursor, Windsurf, Cline, OpenCode, or another supported AI coding client
  • Teams reviewing AI-generated code, MCP tools, prompts, and supply-chain risks
  • Teams needing security evidence before CI, releases, or vendor reviews

Not for

  • Users who only need traditional dependency vulnerability scanning
  • Users looking for a replacement for npm audit
  • Users needing hosted semantic review without configuring a model provider

Required permissions

  • Read access to projects, files, Git diffs, MCP servers, or dependency data being scanned
  • The auto-fix capability may require write access to project files
  • Semantic review uses the model provider selected by the user

Risks and side effects

  • AI-generated code may introduce SQL injection, XSS, command injection, unsafe cryptography, or secrets
  • MCP tools may contain description injection, tool-name spoofing, environment-variable exposure, or command-execution risks
  • AI-generated package names may not exist, creating package-squatting or dependency-confusion risks
  • Semantic review sends relevant content to the selected provider; rule-based scans and MCP scans run locally

Setup

Before you start

Runtime：Node.js

Run npx agent-security-scanner-mcp init claude-code, replacing claude-code with cursor, claude-desktop, windsurf, cline, kilo-code, opencode, or cody. Restart the client afterward. You can also run npx agent-security-scanner-mcp init to use the interactive picker.

Check that it works

After running npx agent-security-scanner-mcp init claude-code (or another client) and restarting, tools such as scan_project, scan_mcp_server, and check_package should appear in the client's tool list; you can also run scanner_health for runtime diagnostics, or verify the CLI with npx agent-security-scanner-mcp scan-project . --verbosity compact.

Troubleshooting

  1. Run doctor to check local setup health
  2. Restart the configured AI client after initialization
  3. Run npx agent-security-scanner-mcp scan-project . --verbosity compact to verify the installation
  4. Use a supported client name: claude-code, cursor, claude-desktop, windsurf, cline, kilo-code, opencode, or cody
  5. Adjust output detail with minimal, compact, or full verbosity

Things to try

Once connected, you can ask your AI assistant things like:

  • Scan this project and give me an A-F security grade
  • Check the imports in ./src/app.ts for hallucinated packages
  • Audit the MCP server at ./path/to/mcp-server before I add it to my agent
  • Scan the changed files between main and HEAD for security issues before I open a PR

Tools 17

scan_security read-only
Scan a single file for security issues.
fix_security writes
Automatically fix security issues that support remediation.
scan_git_diff read-only
Scan Git changes for security issues.
scan_project read-only
Scan a complete project and produce an A-F security grade.
scan_mcp_server read-only
Audit an MCP server before installation for tool and attack risks.
scan_skill read-only
Scan an AI skill for security issues.
scan_agent_prompt read-only
Detect prompt injection and jailbreak attempts.
scan_agent_action read-only
Check shell, file, and network actions for safety before execution.
Show 9 more tools
check_package read-only
Verify whether a package exists in a specified ecosystem.
scan_packages read-only
Verify imports or dependencies in a file to help detect package hallucinations.
sbom_generate writes
Generate a CycloneDX software bill of materials.
sbom_scan_vulnerabilities read-only
Scan dependencies for vulnerabilities using OSV.dev.
sbom_diff read-only
Compare software bill of materials differences.
sbom_export_report writes
Export an SBOM scan report.
get_compliance_controls read-only
Retrieve technical compliance controls for SOC 2, GDPR, or AIUC-1.
evaluate_compliance read-only
Evaluate technical compliance and generate evidence for a project.
scanner_health read-only
Check scanner runtime health.

Use cases

Scan AI-generated project code and diffs
Audit MCP servers before installation
Detect prompt injection and unsafe agent actions
Verify AI-suggested dependencies
Generate SBOMs, scan CVEs, and evaluate compliance

Supported clients

Claude Code
Cursor
Claude Desktop
Windsurf
Cline
Kilo Code
OpenCode
Cody

Listed from the project's documentation, not tested by this site.

Overview

Agent Security Scanner MCP Server is a stdio MCP server that checks code, MCP tools, prompts, and dependencies before an AI agent trusts them. It provides code and repository scanning, MCP server audits, prompt-injection detection, agent-action safety checks, package hallucination detection, SBOM and vulnerability scanning, compliance evidence evaluation, and auto-fix capabilities. Rule-based CLI and MCP scans run locally; semantic review uses the provider selected by the user.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision ef3f20e1de3b Data synced 2026-10-19 Read the FMRS scoring method