Scout Security

What the Scout Security MCP server does

The Scout Security MCP server is presented as a local security scanner for projects. Its native checks cover several recurring application risks: exposed secrets, injection patterns such as string-built SQL or unsafe shell usage, missing security headers, and dependency findings. It can also scan Git history for secrets that were committed and later removed.

The scanner is deterministic and does not call an external AI service. That makes it suitable for local development, automated checks, and agent workflows where results should be repeatable and no source code should be sent to a hosted analysis API. The project is open source under the MIT license and requires no API key, account, or signup.

How it works

The Scout Security MCP server scans a project directory and applies its native detectors. Results can be emitted as a human-readable Markdown report, JSON for downstream tooling, SARIF 2.1.0 for GitHub Code Scanning, or prompts intended for an AI coding assistant. Scout identifies the issue; the developer or their existing AI assistant applies the change.

Scans can act as CI gates. By default, the command exits with status 1 when high or critical findings are present. The threshold can be changed to critical, medium, or report-only mode. A baseline file lets an existing codebase accept current findings and fail only when new findings appear. Finding identity uses the rule, file, and a hash of the flagged line rather than line numbers.

Optional Semgrep and CodeQL engines can be requested and their findings merged with Scout output. These engines are not part of the default scan. Custom YAML rules can add project-specific regular-expression checks with severity, message, file suffix, and optional remediation metadata.

Setup and configuration

The documented zero-install path is uvx scout-security scan ./my-project. A permanent installation uses pip install scout-security. The package also provides a collision-resistant scoutsec executable; the standard scout name may conflict with another package.

Project defaults are read from [tool.scout] in pyproject.toml. Supported settings include excluded paths, selected scanners, external engines, custom rule files, and the default failure threshold. Command-line options override relevant project settings. Paths can also be excluded with repeatable --exclude options.

A pre-commit hook is available, and a GitHub Action wraps installation, scanning, and SARIF upload. The VS Code extension shells out to scoutsec and uses the same project configuration.

Tools and capabilities

The Scout Security MCP server supports:

• Native scanners for secrets, injection, headers, dependencies, and custom rules.
• Markdown, AI-prompt, JSON, and SARIF output formats.
• Severity-based CI failure thresholds.
• Baselines for distinguishing existing findings from new ones.
• Inline suppression comments and scanner- or finding-specific suppression.
• Git-history scanning for secrets in added lines across commits and branches.
• Optional Semgrep and CodeQL result merging.

Limitations and notes

The provided material documents a CLI, pre-commit hook, GitHub Action, and VS Code wrapper, but does not define MCP tools, an MCP configuration block, or a server transport. The Scout Security MCP server should therefore not be assumed to expose a particular MCP command or client configuration from this documentation alone.

The default scan is intentionally dependency-light and deterministic, while Semgrep and CodeQL require separate installations. CodeQL analysis can take minutes because it builds a database. Custom rules are regular-expression checks with metadata; projects needing metavariables or taint analysis should use an optional external engine instead.