surface-audit MCP server

surface-audit

Deterministic security smoke tests for staging, preview, and pre-deploy web apps.

surface-audit sends a small, bounded set of safe probes at a known URL and turns the result into CI-friendly findings, SARIF, Markdown, HTML, and JSON reports. It is designed for teams that want to catch security regressions in headers, TLS, redirects, cookies, CORS, exposed files, and other surface-level mistakes before they promote a build.

⚠️ Authorized use only. This tool sends real HTTP requests to the target. Only run it against systems you own or have explicit written permission to test. See SECURITY.md.

Why Teams Use It

• Safe by default — bounded checks against a known URL, not a crawler and not an exploit framework.
• Regression-oriented — baseline suppression and report diff help answer "what got worse?" instead of just "what exists?"
• CI-native — --fail-on, SARIF, Markdown, HTML, and JSON all fit cleanly into pull requests, code scanning, and release gates.
• LLM-safe MCP support — the optional MCP server exposes only a host allow-listed interface.
• Extensible — checks and renderers are pluggable via entry points.
• Supply-chain aware — PyPI Trusted Publishing, Sigstore signatures, GitHub Releases, and CycloneDX SBOMs are built into the release flow.

Use It When

• you have a staging, preview, or pre-production URL
• you want a deterministic security gate in CI
• you care about security regressions between two deployments
• you want machine-readable output for SARIF, dashboards, or PR comments

Reach for Other Tools When

• you need full crawling or spidering
• you need authenticated scanning workflows
• you want exploit confirmation or a broad template corpus

That positioning is deliberate: surface-audit is strongest as a pre-deploy security smoke test, not as a full DAST platform.

Quick Start

pipx install surface-audit

surface-audit scan https://preview.example.com \
    --scope-host preview.example.com \
    --fail-on HIGH

Detailed installation: docs/INSTALL.md.

Distribution Options

• PyPI — best when you want the CLI in pipx, a virtualenv, or your own Python-based tooling.
• GHCR — best when your CI prefers a prebuilt container image over installing Python dependencies at runtime.
• GitHub Action — best when a preview-environment workflow already lives in GitHub Actions and you want the shortest integration path.

Container-first teams can use the published GHCR image on tagged releases:

docker run --rm ghcr.io/fillbyte/surface-audit:latest \
    scan https://preview.example.com --fail-on HIGH

Use the GitHub Action @v1 tag for the stable major line, or pin an exact action release such as @v1.0.7 when you want fully reproducible workflow inputs. Tagged releases also publish GitHub Release artifacts, CycloneDX SBOMs, and Sigstore signatures.

More end-to-end patterns: docs/RECIPES.md. Release and tag-management details: docs/RELEASE.md.

Security Regression Diff

# Capture a baseline once
surface-audit scan https://preview.example.com \
    --output reports/baseline.json \
    --format json

# Gate only on newly introduced HIGH+ findings
surface-audit scan https://preview.example.com \
    --baseline reports/baseline.json \
    --fail-on HIGH

# Or diff two reports explicitly
surface-audit diff reports/before.json reports/after.json \
    --output reports/diff.json \
    --fail-on-new

GitHub Action

The repository ships an action at the repo root so you can run the scan in a preview-environment workflow without hand-rolling install steps:

- name: Run surface-audit
  uses: fillbyte/surface-audit@v1
  with:
    target: ${{ steps.preview.outputs.url }}
    scope-hosts: preview.example.com
    output: reports/surface-audit.sarif
    format: sarif
    fail-on: HIGH

- name: Upload SARIF
  uses: github/codeql-action/upload-sarif@v4
  with:
    sarif_file: reports/surface-audit.sarif

More end-to-end patterns: docs/RECIPES.md.

Sample Output

Target: https://preview.example.com/
Summary: HIGH 1  MEDIUM 2  LOW 1

HIGH     security-headers        Missing Content-Security-Policy header
MEDIUM   auth-cookies            Cookie 'sessionid' missing SameSite
MEDIUM   security-txt            Missing /.well-known/security.txt
LOW      directory-listing       Auto-generated index page exposed

Built-in checks

Check ID	OWASP	Summary
security-headers	A05	Missing and weakly-configured HSTS / CSP / XFO / Referrer / Permissions
ssl-tls	A02	Weak ciphers or obsolete TLS versions
https-redirect	A02	Plain HTTP does not redirect to HTTPS
cross-origin-isolation	A05	Missing COOP / COEP / CORP isolation headers
xss-reflection	A03	Query-string input reflected without output encoding
sql-injection	A03	Database error messages leaked on meta-character input
csrf	A01	Mutating HTML forms without a recognizable anti-CSRF token
auth-cookies	A07	Cookies missing Secure / HttpOnly / SameSite
open-redirect	A01	Query parameters that allow off-origin 30x redirects
misconfiguration	A05	Well-known exposed paths (.env, .git, admin consoles)
directory-listing	A05	Auto-generated directory index pages
cors	A05	Permissive CORS reflections and wildcard-with-credentials
security-txt	A09	Missing /.well-known/security.txt (RFC 9116)

Run surface-audit list-checks to see what is registered in your environment (including any third-party plugins).

Usage

# Save a JSON report and fail CI on HIGH+ findings
surface-audit scan https://example.com \
    --output reports/example.json --format json --fail-on HIGH

# Emit SARIF for GitHub Advanced Security
surface-audit scan https://example.com \
    --output reports/example.sarif --format sarif

# Run only a subset of checks
surface-audit scan https://example.com \
    --enable security-headers --enable ssl-tls

Full CLI and library reference: docs/USAGE.md.

MCP integration

pip install "surface-audit[mcp]"
surface-audit mcp-serve --allow-host staging.example.com

This exposes scan, list_checks, list_formats, and render_report tools to local MCP clients while keeping scans gated behind an explicit host allow-list.

Project Site

The project site highlights the smoke-test workflow, GitHub Action, sample artifacts, and core adoption patterns:

• Project landing page
• Console sample
• HTML sample
• SARIF sample

Download trends remain available via PyPI Stats.

Library example

import asyncio
from surface_audit import Scanner, ScannerConfig


async def main() -> None:
    report = await Scanner(
        "https://example.com",
        config=ScannerConfig(max_concurrency=4, timeout=5.0),
    ).run()
    for finding in report.findings:
        print(finding.severity.value, finding.check_id, finding.title)


asyncio.run(main())

Extend it

Any package shipping an entry point under surface_audit.checks adds a check. A starter template for third-party checks lives under examples/plugin-template:

# your-plugin/pyproject.toml
[project.entry-points."surface_audit.checks"]
cors_wildcard = "my_checks.cors:PermissiveCORSCheck"

See docs/ARCHITECTURE.md for layering, SOLID scorecard, and design patterns.

Documentation

• docs/INSTALL.md — per-platform installation
• docs/USAGE.md — CLI, library, and CI recipes
• docs/RECIPES.md — preview, SARIF, baseline, MCP, and action recipes
• docs/RELEASE.md — PyPI, GHCR, GitHub Release, and v1 tag process
• docs/ARCHITECTURE.md — layering and extension points
• docs/SCHEMA.md — JSON report contract
• examples/plugin-template — starter template for third-party checks
• SUPPORT.md — where to ask questions and report the right thing
• CONTRIBUTING.md — development workflow
• SECURITY.md — vulnerability disclosure
• CODE_OF_CONDUCT.md — community expectations

Development

make install   # set up .venv with dev extras
make all       # ruff + mypy + bandit + pytest

CI runs on every push across Python 3.10 – 3.14 — see .github/workflows/ci.yml.

License

Apache License 2.0 — Copyright © 2026 Uğur Kontel. See also NOTICE for attribution requirements that apply to redistributions.

---

surface-audit is an open-source project from Fillbyte.