# Security Audit Report — mcp_snowflake_server

- **Report ID:** `452b183a-bc9b-4a05-bfc1-3950515334e9`
- **Generated:** 2026-07-21T23:27:34.239621+00:00
- **Signature:** unsigned (cosign keyless signing runs in CI; Req 22.4)

## 1. Executive Summary

**Badge:** Verified (composite)  
**Security score:** 90.55

| Scanner | Badge |
| --- | --- |
| agent-audit-kit | Unsafe |
| agentshield | unavailable |
| bearer | Verified |
| cisco-skill-scanner | Verified |
| nerlo-behavioral | Verified |
| nerlo-install-instruction | Unsafe |
| osv-scanner | unavailable |
| trivy | Verified |

| Severity | Findings |
| --- | --- |
| critical | 0 |
| high | 3 |
| medium | 3 |
| low | 4 |

mcp_snowflake_server may be integrated with compensating controls: review the 3 high-severity findings below, restrict granted permissions to the minimum the manifest declares, and subscribe to monitoring alerts for score changes.

## 2. Source Provenance

- **Repository:** https://github.com/isaacwasserman/mcp-snowflake-server
- **Commit scanned:** `unknown`
- **License:** unknown
- **Maintainer:** unknown
- **Version:** 0.4.0

## 3. Per-Scanner Detail

### agentshield (v1.4.0) — unavailable / n/a

No findings.

### cisco-skill-scanner (v2.0.11) — Verified / 93.5

- **[low] Dangerous data flow in command pipeline** — Pipeline downloads data from the network and executes it: 'curl -LsSf https://astral.sh/uv/install.sh \| sh'. This is a remote code execution pattern. (Note: uses a well-known installer URL - likely a standard installation command.) (/repo/README.md:2)
- **[low] Vague skill description** — Skill description is too short (16 chars). Provide detailed explanation. (/repo/SKILL.md:None)
- **[informational] Skill does not specify a license** — Skill manifest does not include a 'license' field. Specifying a license helps users understand usage terms. (/repo/SKILL.md:None)
- **[low] Dangerous data flow in command pipeline** — Pipeline downloads data from the network and executes it: 'curl -LsSf https://astral.sh/uv/install.sh \| sh'. This is a remote code execution pattern. (Note: uses a well-known installer URL - likely a standard installation command.) (/repo/SKILL.md:2)

### agent-audit-kit (v0.3.26) — Unsafe / 53.0

- **[high] Generic high-entropy secret** — A value assigned to a secret-like key has high Shannon entropy, indicating a likely credential or API key. (src/mcp_snowflake_server/__init__.py:159)
- **[medium] No lockfile present** — A package manifest exists but no lockfile was found. Without lockfiles, dependency versions float and can be silently updated. (pyproject.toml:None)
- **[low] MCP server repo missing SECURITY.md or security_contact** — A repository whose name or pyproject keywords declare it as an MCP server ships without a top-level SECURITY.md AND without a 'security_contact' entry in marketplace.json / pyproject.toml / package.json. Anthropic's April 2026 SECURITY.md guidance makes this the baseline expectation so researchers have a channel. (SECURITY.md:None)
- **[medium] Repo depends on a third-party agent-platform SDK** — The project depends on an agent-platform SDK (context-ai, langsmith, helicone, langfuse, humanloop, MCP SDK). Informational finding so reviewers audit the vendor's OAuth-scope footprint before merging. Raised to MEDIUM because the April 19 2026 Vercel × Context.ai incident showed a single vendor compromise can turn into a production breach via transitive OAuth grants. (pyproject.toml:2)
- **[high] Vulnerable MCP SDK version pinned (DNS-rebinding fix missing)** — A project dependency manifest (requirements.txt, pyproject.toml, package.json, pom.xml, build.gradle) pins an MCP SDK at a version below the DNS-rebinding fix. Patched versions: Python 'mcp' >= 1.23.0, TS '@modelcontextprotocol/sdk' >= 1.21.1, Java 'io.modelcontextprotocol.sdk:mcp-core' >= 0.11.0, '@apollo/mcp-server' >= 1.7.0. Even if the project never serves over StreamableHTTP itself, transitive servers built on the SDK inherit the bug. (pyproject.toml:8)
- **[medium] Third-party GitHub Action not pinned by full commit SHA** — A workflow in '.github/workflows/' uses a third-party Action ('owner/action@ref') where 'ref' is a tag or branch name instead of a 40-character commit SHA. A repo-takeover of the Action's publisher can re-point the tag to a malicious revision — the downstream repo consuming it will happily run the new code with 'GITHUB_TOKEN' and write permissions. GitHub's April 2026 Security Roadmap ships Immutable Actions and makes SHA pinning the default policy. (.github/workflows/python-publish.yml:71)

### bearer (v2.0.2) — Verified / 100.0

No findings.

### nerlo-behavioral (v0.1.0) — Verified / 100.0

No findings.

### nerlo-install-instruction (v0.1.0) — Unsafe / 59.0

- **[high] opt.nerlo-rules.nerlo-install-pipe-to-shell** — Documentation instructs piping a downloaded artifact directly into a shell (curl/wget/iwr ... \| bash/sh/iex). This is the curl-pipe-bash install shape used by the ClawHub skill droppers and countless malware loaders — the fetched payload is executed with zero inspection. Confirm this is expected; it is a strong weaponized-documentation signal. (/repo/README.md:167)

### trivy (v0.71.0) — Verified / 100.0

No findings.

### osv-scanner (v2.3.8) — unavailable / n/a

No findings.

## 4. Threat Model

Threat model synthesis has not yet run for this scan. This section is generated by the registry's LLM pipeline (Req 22.3) and will appear in the next regeneration of this report.

## 5. Audit Chain

- **Scan job:** `9de564ce-e5b3-4131-afc8-fc05ecd8404f`
- **Completed:** 2026-07-21T02:54:24.505706+00:00
- **Scanner base image:** `us-central1-docker.pkg.dev/nerlo-vsk-prod/nerlo/scanner-base@sha256:e6a801e1d66ee5b43f7cfc10378bf8794329e99209e3c5a76133ec3f4984b3a6`
- **AI decision log entries:** 1
  - `fbdd9997-bb32-4ebf-a661-69c7c25ca8f9`

## 6. Appendix — Raw Scanner Output

```json
[
  {
    "scanner_name": "agentshield",
    "scanner_version": "1.4.0",
    "score": 100.0,
    "scanner_badge": "Verified",
    "findings": [],
    "execution_duration_seconds": 1.5405062229983741,
    "status": "not_applicable",
    "metadata": {
      "source": "npm",
      "source_url": "https://www.npmjs.com/package/ecc-agentshield",
      "install_command": "npm install -g ecc-agentshield@1.4.0",
      "scans_performed": []
    },
    "display_score": null,
    "display_badge": "unavailable"
  },
  {
    "scanner_name": "cisco-skill-scanner",
    "scanner_version": "2.0.11",
    "score": 93.5,
    "scanner_badge": "Verified",
    "findings": [
      {
        "tool_name": "cisco-skill-scanner",
        "severity": "low",
        "category": "command_injection",
        "file_path": "/repo/README.md",
        "line_number": 2,
        "rule_identifier": "PIPELINE_TAINT_FLOW",
        "title": "Dangerous data flow in command pipeline",
        "description": "Pipeline downloads data from the network and executes it: `curl -LsSf https://astral.sh/uv/install.sh | sh`. This is a remote code execution pattern. (Note: uses a well-known installer URL - likely a standard installation command.)",
        "remediation": "Review the command pipeline. Avoid piping sensitive data to network commands or shell execution."
      },
      {
        "tool_name": "cisco-skill-scanner",
        "severity": "low",
        "category": "social_engineering",
        "file_path": "/repo/SKILL.md",
        "line_number": null,
        "rule_identifier": "SOCIAL_ENG_VAGUE_DESCRIPTION",
        "title": "Vague skill description",
        "description": "Skill description is too short (16 chars). Provide detailed explanation.",
        "remediation": "Provide a clear, detailed description of what the skill does and when to use it"
      },
      {
        "tool_name": "cisco-skill-scanner",
        "severity": "informational",
        "category": "policy_violation",
        "file_path": "/repo/SKILL.md",
        "line_number": null,
        "rule_identifier": "MANIFEST_MISSING_LICENSE",
        "title": "Skill does not specify a license",
        "description": "Skill manifest does not include a 'license' field. Specifying a license helps users understand usage terms.",
        "remediation": "Add 'license' field to SKILL.md frontmatter (e.g., MIT, Apache-2.0)"
      },
      {
        "tool_name": "cisco-skill-scanner",
        "severity": "low",
        "category": "command_injection",
        "file_path": "/repo/SKILL.md",
        "line_number": 2,
        "rule_identifier": "PIPELINE_TAINT_FLOW",
        "title": "Dangerous data flow in command pipeline",
        "description": "Pipeline downloads data from the network and executes it: `curl -LsSf https://astral.sh/uv/install.sh | sh`. This is a remote code execution pattern. (Note: uses a well-known installer URL - likely a standard installation command.)",
        "remediation": "Review the command pipeline. Avoid piping sensitive data to network commands or shell execution."
      }
    ],
    "execution_duration_seconds": 28.57257441400725,
    "status": "complete",
    "metadata": {
      "source": "pypi",
      "source_url": "https://pypi.org/project/cisco-ai-skill-scanner/2.0.11/",
      "report_type": "cisco-skill-sast",
      "analyzers_used": [
        "bytecode",
        "pipeline",
        "static_analyzer"
      ],
      "skills_scanned": [
        "repo"
      ],
      "install_command": "pip install --require-hashes -r docker/scanner-base/cisco-skill-scanner/requirements.txt",
      "severity_counts": {
        "low": 3,
        "high": 0,
        "medium": 0,
        "critical": 0,
        "informational": 1
      }
    },
    "display_score": 93.5,
    "display_badge": "Verified"
  },
  {
    "scanner_name": "agent-audit-kit",
    "scanner_version": "0.3.26",
    "score": 53.0,
    "scanner_badge": "Unsafe",
    "findings": [
      {
        "tool_name": "agent-audit-kit",
        "severity": "high",
        "category": "secret-exposure",
        "file_path": "src/mcp_snowflake_server/__init__.py",
        "line_number": 159,
        "rule_identifier": "AAK-SECRET-004",
        "title": "Generic high-entropy secret",
        "description": "A value assigned to a secret-like key has high Shannon entropy, indicating a likely credential or API key.",
        "remediation": "Move to environment variables or secrets manager."
      },
      {
        "tool_name": "agent-audit-kit",
        "severity": "medium",
        "category": "supply-chain",
        "file_path": "pyproject.toml",
        "line_number": null,
        "rule_identifier": "AAK-SUPPLY-004",
        "title": "No lockfile present",
        "description": "A package manifest exists but no lockfile was found. Without lockfiles, dependency versions float and can be silently updated.",
        "remediation": "Generate and commit lockfile."
      },
      {
        "tool_name": "agent-audit-kit",
        "severity": "low",
        "category": "supply-chain",
        "file_path": "SECURITY.md",
        "line_number": null,
        "rule_identifier": "AAK-SEC-MD-001",
        "title": "MCP server repo missing SECURITY.md or security_contact",
        "description": "A repository whose name or pyproject keywords declare it as an MCP server ships without a top-level SECURITY.md AND without a `security_contact` entry in marketplace.json / pyproject.toml / package.json. Anthropic's April 2026 SECURITY.md guidance makes this the baseline expectation so researchers have a channel.",
        "remediation": "Add SECURITY.md at the repo root with a disclosure email and response SLA; OR add `security_contact` to the project manifest."
      },
      {
        "tool_name": "agent-audit-kit",
        "severity": "medium",
        "category": "supply-chain",
        "file_path": "pyproject.toml",
        "line_number": 2,
        "rule_identifier": "AAK-OAUTH-3P-001",
        "title": "Repo depends on a third-party agent-platform SDK",
        "description": "The project depends on an agent-platform SDK (context-ai, langsmith, helicone, langfuse, humanloop, MCP SDK). Informational finding so reviewers audit the vendor's OAuth-scope footprint before merging. Raised to MEDIUM because the April 19 2026 Vercel \u00d7 Context.ai incident showed a single vendor compromise can turn into a production breach via transitive OAuth grants.",
        "remediation": "Pin the SDK to an exact version, audit the OAuth scopes it requests, and keep any deployment-level grants (Vercel, GCP, Workspace) in a secrets vault \u2014 never in a committed env file. See Vercel's bulletin for sensitive-env-var guidance: https://vercel.com/kb/bulletin/vercel-april-2026-security-incident"
      },
      {
        "tool_name": "agent-audit-kit",
        "severity": "high",
        "category": "supply-chain",
        "file_path": "pyproject.toml",
        "line_number": 8,
        "rule_identifier": "AAK-DNS-REBIND-002",
        "title": "Vulnerable MCP SDK version pinned (DNS-rebinding fix missing)",
        "description": "A project dependency manifest (requirements.txt, pyproject.toml, package.json, pom.xml, build.gradle) pins an MCP SDK at a version below the DNS-rebinding fix. Patched versions: Python `mcp` >= 1.23.0, TS `@modelcontextprotocol/sdk` >= 1.21.1, Java `io.modelcontextprotocol.sdk:mcp-core` >= 0.11.0, `@apollo/mcp-server` >= 1.7.0. Even if the project never serves over StreamableHTTP itself, transitive servers built on the SDK inherit the bug.",
        "remediation": "Bump the SDK to the patched version listed in the rule title. If a bump is not yet possible, ensure every transport surface has its own Host-header allow-list (see AAK-DNS-REBIND-001 remediation)."
      },
      {
        "tool_name": "agent-audit-kit",
        "severity": "medium",
        "category": "supply-chain",
        "file_path": ".github/workflows/python-publish.yml",
        "line_number": 71,
        "rule_identifier": "AAK-GHA-IMMUTABLE-001",
        "title": "Third-party GitHub Action not pinned by full commit SHA",
        "description": "A workflow in `.github/workflows/` uses a third-party Action (`owner/action@ref`) where `ref` is a tag or branch name instead of a 40-character commit SHA. A repo-takeover of the Action's publisher can re-point the tag to a malicious revision \u2014 the downstream repo consuming it will happily run the new code with `GITHUB_TOKEN` and write permissions. GitHub's April 2026 Security Roadmap ships Immutable Actions and makes SHA pinning the default policy.",
        "remediation": "Repin third-party Actions to a 40-character commit SHA and add a `# v1.2.3`-style trailing comment for humans. First-party Actions under `actions/` and `github/` are exempt (they now ship Immutable Actions). Dependabot will auto-bump SHA pins when `update-type: all` is set."
      }
    ],
    "execution_duration_seconds": 9.031663877001847,
    "status": "complete",
    "metadata": {
      "source": "pypi",
      "source_url": "https://pypi.org/project/agent-audit-kit/0.3.26/",
      "report_type": "agent-audit-kit-sast",
      "files_scanned": 12,
      "install_command": "pip install --require-hashes -r docker/scanner-base/agent-audit-kit/requirements.txt",
      "rules_evaluated": 211,
      "severity_counts": {
        "low": 1,
        "high": 2,
        "medium": 3,
        "critical": 0,
        "informational": 0
      }
    },
    "display_score": 53.0,
    "display_badge": "Unsafe"
  },
  {
    "scanner_name": "bearer",
    "scanner_version": "2.0.2",
    "score": 100.0,
    "scanner_badge": "Verified",
    "findings": [],
    "execution_duration_seconds": 9.887080272004823,
    "status": "complete",
    "metadata": {
      "source": "github-releases",
      "source_url": "https://github.com/Bearer/bearer",
      "report_type": "security",
      "rules_loaded": 1,
      "install_command": "curl -sfL https://raw.githubusercontent.com/Bearer/bearer/main/contrib/install.sh | sh -s -- -b /usr/local/bin \"v2.0.2\""
    },
    "display_score": 100.0,
    "display_badge": "Verified"
  },
  {
    "scanner_name": "nerlo-behavioral",
    "scanner_version": "0.1.0",
    "score": 100.0,
    "scanner_badge": "Verified",
    "findings": [],
    "execution_duration_seconds": 40.97404797100171,
    "status": "complete",
    "metadata": {
      "source": "nerlo-original",
      "source_url": "https://github.com/nerlo-ai/nerlo",
      "report_type": "nerlo-behavioral",
      "ruleset_path": "/opt/nerlo-rules/exfiltration.yaml",
      "files_scanned": 5,
      "install_command": "pip install 'semgrep==1.97.0'"
    },
    "display_score": 100.0,
    "display_badge": "Verified"
  },
  {
    "scanner_name": "nerlo-install-instruction",
    "scanner_version": "0.1.0",
    "score": 59.0,
    "scanner_badge": "Unsafe",
    "findings": [
      {
        "tool_name": "nerlo-install-instruction",
        "severity": "high",
        "category": "install-instruction-exec",
        "file_path": "/repo/README.md",
        "line_number": 167,
        "rule_identifier": "opt.nerlo-rules.nerlo-install-pipe-to-shell",
        "title": "opt.nerlo-rules.nerlo-install-pipe-to-shell",
        "description": "Documentation instructs piping a downloaded artifact directly into a shell (curl/wget/iwr ... | bash/sh/iex). This is the curl-pipe-bash install shape used by the ClawHub skill droppers and countless malware loaders \u2014 the fetched payload is executed with zero inspection. Confirm this is expected; it is a strong weaponized-documentation signal.",
        "remediation": null
      }
    ],
    "execution_duration_seconds": 40.7502596740087,
    "status": "complete",
    "metadata": {
      "source": "nerlo-original",
      "source_url": "https://github.com/nerlo-ai/nerlo",
      "report_type": "nerlo-install-instruction",
      "ruleset_path": "/opt/nerlo-rules/install_instructions.yaml",
      "files_scanned": 1,
      "install_command": "pip install 'semgrep==1.97.0'",
      "severity_counts": {
        "low": 0,
        "high": 1,
        "medium": 0,
        "critical": 0,
        "informational": 0
      }
    },
    "display_score": 59.0,
    "display_badge": "Unsafe"
  },
  {
    "scanner_name": "trivy",
    "scanner_version": "0.71.0",
    "score": 100.0,
    "scanner_badge": "Verified",
    "findings": [],
    "execution_duration_seconds": 1.0391425209963927,
    "status": "complete",
    "metadata": {
      "source": "github-releases",
      "source_url": "https://github.com/aquasecurity/trivy/releases/tag/v0.71.0",
      "report_type": "filesystem-vulnerability",
      "install_command": "curl -sfL -o /tmp/trivy.deb https://github.com/aquasecurity/trivy/releases/download/v0.71.0/trivy_0.71.0_Linux-64bit.deb && echo '<sha256>  /tmp/trivy.deb' | sha256sum -c - && dpkg -i /tmp/trivy.deb",
      "manifests_scanned": []
    },
    "display_score": 100.0,
    "display_badge": "Verified"
  },
  {
    "scanner_name": "osv-scanner",
    "scanner_version": "2.3.8",
    "score": 100.0,
    "scanner_badge": "Verified",
    "findings": [],
    "execution_duration_seconds": 0.7118588519952027,
    "status": "not_applicable",
    "metadata": {
      "source": "github-releases",
      "source_url": "https://github.com/google/osv-scanner/releases/tag/v2.3.8",
      "report_type": "osv-vulnerability",
      "ecosystems_seen": [],
      "install_command": "curl -sfL -o /usr/local/bin/osv-scanner https://github.com/google/osv-scanner/releases/download/v2.3.8/osv-scanner_linux_amd64 && echo '<sha256>  /usr/local/bin/osv-scanner' | sha256sum -c - && chmod +x /usr/local/bin/osv-scanner",
      "manifests_scanned": [],
      "finding_id_aliases": {},
      "cross_scanner_correlation": {
        "only_osv": [],
        "only_trivy": [],
        "intersection_ids": []
      }
    },
    "display_score": null,
    "display_badge": "unavailable"
  }
]
```