Config Drift Guard MCP

A read-only Model Context Protocol server for comparing environment configuration without exposing detected secrets.

Capabilities

• Parse .env, YAML, JSON, and Java Properties files.

• Normalize nested configuration into stable key paths.

• Detect added, removed, type-changed, value-changed, and secret-changed entries.

• Mask likely credentials, private keys, bearer tokens, and credential-bearing URLs.

• Restrict file access to explicitly allowed directories.

• Run entirely locally without network requests.

Related MCP server: vault-kv-mcp

Tools

inspect_config

Parse one file and return normalized entries, value types, warnings, and protected values.

compare_configs

Compare a baseline file with a target file and return drift items with risk levels.

scan_secrets

Find likely plaintext secrets while returning only masks and SHA-256 fingerprints.

Install

npm install
npm run build

Node.js 20 or later is required.

Configure

Set CONFIG_DRIFT_ALLOWED_ROOTS to one or more directories. Use the operating system path delimiter (; on Windows, : on Linux and macOS).

# Windows PowerShell
$env:CONFIG_DRIFT_ALLOWED_ROOTS='D:\configs\dev;D:\configs\prod'
node dist/index.js

# Linux or macOS
CONFIG_DRIFT_ALLOWED_ROOTS=/srv/config/dev:/srv/config/prod node dist/index.js

If the variable is omitted, only the current working directory is allowed.

MCP Client Configuration

{
  "mcpServers": {
    "config-drift-guard": {
      "command": "node",
      "args": ["/absolute/path/config-drift-guard-mcp/dist/index.js"],
      "env": {
        "CONFIG_DRIFT_ALLOWED_ROOTS": "/absolute/path/configs"
      }
    }
  }
}

Example

The repository includes examples/dev.yaml and examples/prod.yaml.

compare_configs({
  "baselineFile": "/absolute/path/examples/dev.yaml",
  "targetFile": "/absolute/path/examples/prod.yaml"
})

The result distinguishes missing keys, type drift, ordinary value drift, and secret fingerprint changes. Detected secret plaintext is never returned.

Paired Skill

skills/config-drift-remediation converts MCP evidence into:

• Expected-environment differences.

• Prioritized defects.

• Secret-rotation recommendations.

• Key-level patch guidance.

• A post-change verification checklist.

Security Model

• Read-only tools.

• No network access.

• No expression or script execution.

• Maximum input file size of 5 MiB.

• Allowed-directory enforcement.

• Secret masks and one-way fingerprints only.

• No configuration plaintext logging.

Development

npm test
npm run build
npm pack --dry-run

License

MIT