# Agent Bom

> Open security scanner for agentic infrastructure — agents, MCP, packages, blast radius, runtime, and trust across MCP servers, skills, packages, and agents in Cortex Code.

- Skill: `msaad00/agent-bom-2` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add msaad00/agent-bom-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/msaad00/agent-bom-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: msaad00 (https://skillmd.com/u/msaad00)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/msaad00/agent-bom-2

---


# agent-bom — Security Platform for Agentic Infrastructure

Scan your MCP servers, AI agents, and their packages: CVEs and supply-chain
risk (malicious/typosquat) in the packages, plus credential exposure,
identity/auth/access posture, and MCP config risk on the servers and agents.
Maps blast radius from a finding to the credentials and tools it can reach.

## When to Use

- Before installing a new MCP server or skill
- To audit your Cortex Code MCP configuration for vulnerabilities
- To check if a specific package has known CVEs
- To map the blast radius of a vulnerability across your agent setup
- To generate an SBOM for compliance (CycloneDX, SPDX)
- To verify package integrity via Sigstore provenance

## What This Skill Provides

- **CVE scanning** with enrichment from OSV, NVD, EPSS, CISA KEV
- **MCP server discovery** across real AI developer tools, including Cortex Code
- **Blast radius mapping** — which agents, credentials, and tools are exposed
- **Compliance mapping** across OWASP, NIST, MITRE, EU AI Act, and related frameworks
- **Runtime proxy** for MCP traffic interception and policy enforcement
- **SKILL.md trust assessment** with 17 behavioral risk patterns

## Install

```bash
# Install via pip or pipx
pipx install agent-bom

# Or run directly with uvx (no install needed)
uvx agent-bom scan
```

## Instructions

### Scan the current MCP setup

```bash
agent-bom scan
```

Auto-discovers all configured MCP clients (Cortex Code, Claude Desktop, Cursor, etc.),
resolves their server packages, and scans for CVEs.

### Check a package before installing

```bash
agent-bom check <package>@<version> --ecosystem <npm|pypi|go|maven|cargo|nuget|gem>
```

### Map blast radius of a CVE

```bash
agent-bom scan --enrich
# Then review the blast radius section in the output
```

### Generate SBOM

```bash
agent-bom scan -f cyclonedx -o sbom.json
agent-bom scan -f spdx -o sbom.spdx.json
```

### Scan with GPU infrastructure detection

```bash
agent-bom scan --gpu-scan
```

### Run as MCP server (for other AI tools)

```bash
agent-bom mcp server
```

### Proxy MCP traffic with policy enforcement

```bash
agent-bom proxy "npx @modelcontextprotocol/server-filesystem /workspace" --policy policy.yml
```

## Common Patterns

### Pre-install security check
When a user asks to install a new MCP server or package:
1. Run `agent-bom check <package> --ecosystem <ecosystem>`
2. If vulnerabilities found, show severity and suggest fixed versions
3. If clean, confirm safe to proceed

### Full security audit
When a user asks for a security review:
1. Run `agent-bom scan --enrich`
2. Summarize findings by severity
3. Highlight any credential exposure
4. Show blast radius for critical/high findings

### Compliance report
When compliance documentation is needed:
1. Run `agent-bom scan -f cyclonedx -o sbom.json`
2. Run `agent-bom scan --enrich --compliance` for compliance framework mapping
3. Report maps to OWASP, NIST, EU AI Act, ISO 27001, SOC 2

## Best Practices

- Always run `agent-bom check` before adding new MCP servers to your `mcp.json`
- Use `--enrich` for full NVD/EPSS/KEV enrichment (slower but more complete)
- Review blast radius for any CRITICAL or HIGH findings
- Use `agent-bom scan -f sarif` for CI/CD integration with GitHub Security tab
- No API keys required for basic scanning (NVD_API_KEY optional for higher rate limits)

## Privacy

- Only public package names and versions are sent to vulnerability databases
- Credentials in MCP configs are redacted by `sanitize_env_vars()`
- No telemetry, no tracking, no analytics
- Apache 2.0 licensed, fully open source