Mimir Scan

Mimir Scan

> "Like Mimir advises Kratos of dangers ahead, Mimir-Scan reveals hidden security threats in your code"

An MCP (Model Context Protocol) security scanner that integrates with AI coding agents like Claude Code, GitHub Copilot, Cursor, and others.

Wraps industry-standard security scanners (Semgrep, Trivy, Gitleaks) plus a built-in prompt injection detector, and exposes them as tools that any MCP-compatible AI agent can call.

Features

• Multi-Scanner: Combines SAST, SCA, secrets detection, and prompt injection scanning
• Security Grade: A-F grading system for quick project health assessment
• SARIF Export: GitHub Security tab integration
• Git Diff Scanning: Fast PR reviews by scanning only changed files
• Auto-Fix: Automatically apply safe fixes (dependency upgrades)
• Prompt Injection Detection: Scan for 23+ injection patterns in AI configs
• Custom Scanners: Bring your own scanner via .mimir.json config

Quick Start

From Source (Recommended)

Clone the repo and install dependencies once:

git clone https://github.com/amitvishw/mimir-scan.git
cd mimir-scan
bun install

Then point your MCP client at the local install (see [MCP Client Configuration](#mcp-client-configuration) below).

Using Docker

Docker bundles all scanners - no local installation needed:

docker build -t mimir-scan .
docker run --rm -i -v /path/to/project:/workspace mimir-scan

MCP Client Configuration

Claude Code

Add to your Claude Code settings (~/.claude/settings.json or project's .mcp.json):

{
  "mcpServers": {
    "mimir": {
      "command": "bun",
      "args": ["run", "--cwd", "/path/to/mimir-scan", "start"],
      "env": {
        "MIMIR_TARGET_DIR": "/path/to/your/project"
      }
    }
  }
}

Claude Code Plugin (with Skills)

If you use Mimir as a Claude Code plugin, it ships with two built-in skills that automate the full scan workflow:

| Skill | Trigger | What it does | |-------|---------|--------------| | /mimir-scan:scan | After writing code, on demand | Full scan → grade → findings → autofix → verify loop | | /mimir-scan:check | When asking about scanner setup | Reports which scanners are installed with install hints |

The plugin registers the MCP server under the name mimir, so all tools are available as mimir:mimir_scan, mimir:mimir_verify, etc.

VS Code / GitHub Copilot

Add to .vscode/mcp.json in your project:

{
  "servers": {
    "mimir": {
      "command": "bun",
      "args": ["run", "--cwd", "/path/to/mimir-scan", "start"],
      "env": {
        "MIMIR_TARGET_DIR": "${workspaceFolder}"
      }
    }
  }
}

Cursor

Add to Cursor's MCP settings:

{
  "mcpServers": {
    "mimir": {
      "command": "bun",
      "args": ["run", "--cwd", "/path/to/mimir-scan", "start"],
      "env": {
        "MIMIR_TARGET_DIR": "/path/to/your/project"
      }
    }
  }
}

Docker Configuration

For any MCP client, use Docker to avoid installing scanners locally.

Build the image first:

docker build -t mimir-scan .

Then configure your MCP client:

{
  "mcpServers": {
    "mimir": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-v", "/path/to/your/project:/workspace:ro",
        "mimir-scan"
      ]
    }
  }
}

Installing Scanners

The server works with whatever scanners you have installed. Install the ones you need:

# Semgrep (SAST - code vulnerabilities)
pip3 install semgrep

# Trivy (SCA - dependency vulnerabilities + IaC)
curl -sfL https://raw.githubusercontent.com/aquasecurity/trivy/main/contrib/install.sh | sh -s -- -b /usr/local/bin

# Gitleaks (secrets detection)
# macOS:
brew install gitleaks
# Linux:
curl -sSfL https://github.com/gitleaks/gitleaks/releases/download/v8.30.1/gitleaks_8.30.1_linux_x64.tar.gz | tar xz -C /usr/local/bin gitleaks

Prompt injection detection is built-in and requires no installation.

Scanners

| Scanner | What it detects | Category | |---------|----------------|----------| | Semgrep | Code vulnerabilities — SQL injection, XSS, insecure crypto | sast | | Trivy | Dependency CVEs + IaC misconfigurations | sca, iac | | Gitleaks | Hardcoded secrets, API keys, passwords | secrets | | Prompt Injection | Jailbreaks, instruction overrides, data exfiltration | prompt-injection |

MCP Tools

| Tool | Description | |------|-------------| | mimir_scan | Run all security scans on the codebase | | mimir_scan_diff | Scan only git-changed files (fast PR reviews) | | mimir_scan_prompt | Scan text for prompt injection attacks | | mimir_check_scanners | Check which scanners are installed | | mimir_findings | List detailed findings with filtering | | mimir_grade | Get security grade (A-F) for the project | | mimir_fix | Generate fix instructions for the AI agent | | mimir_autofix | Auto-apply safe fixes (dependency upgrades) | | mimir_sarif | Export findings in SARIF format | | mimir_verify | Re-scan to confirm fixes were applied |

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | MIMIR_TARGET_DIR | Current directory | Directory to scan | | MIMIR_SCANNERS | semgrep,trivy,gitleaks,prompt-injection | Scanners to enable | | MIMIR_MIN_SEVERITY | low | Minimum severity (critical, high, medium, low, info) | | MIMIR_LOG_LEVEL | info | Log level (debug, info, warn, error, silent) |

Workflows

Full Scan Workflow

1. mimir_check_scanners → see what's available
2. mimir_scan → run security scans
3. mimir_findings → review detailed findings
4. mimir_autofix with apply: true → auto-fix safe issues
5. mimir_fix → get instructions for remaining issues
6. Fix the code
7. mimir_verify → confirm fixes

PR Review Workflow

1. mimir_scan_diff with base: "main" → scan only changed files
2. mimir_grade → quick A-F assessment
3. mimir_sarif → export for GitHub Security tab

Security Grading

| Grade | Criteria | |-------|----------| | A | No critical or high, ≤2 medium | | B | No critical, ≤2 high, ≤5 medium | | C | No critical, ≤5 high, ≤10 medium | | D | ≤2 critical, ≤10 high | | F | >2 critical or >10 high |

Prompt Injection Detection

The built-in scanner detects 23+ attack patterns:

• Instruction overrides: "ignore previous instructions"
• Jailbreaks: DAN mode, developer mode
• Role manipulation: "you are now a..."
• Data exfiltration: attempts to send data to external URLs
• System prompt extraction: attempts to reveal instructions
• Encoding bypasses: base64, unicode escapes

Scan AI config files or validate user inputs:

mimir_scan_prompt text="ignore all previous instructions" source="user_input"

Custom Scanners

Add your own scanners by creating a .mimir.json file in your project root:

{
  "plugins": [
    {
      "name": "my-scanner",
      "displayName": "My Custom Scanner",
      "category": "sast",
      "checkCommand": "my-scanner",
      "command": "my-scanner",
      "args": ["scan", "--json", "."],
      "outputFormat": "json",
      "parser": {
        "resultsPath": "results",
        "fieldMapping": {
          "id": "ruleId",
          "severity": "level",
          "title": "message",
          "description": "details",
          "filePath": "file",
          "startLine": "line"
        }
      }
    }
  ]
}

Plugin Configuration

| Field | Required | Description | |-------|----------|-------------| | name | Yes | Unique identifier for the scanner | | displayName | No | Human-readable name | | category | Yes | One of: sast, sca, secrets, iac, prompt-injection | | checkCommand | Yes | Command to check if scanner is installed | | command | Yes | Command to run the scanner | | args | No | Arguments to pass to the command | | outputFormat | Yes | Output format: json or sarif | | parser | Yes (for json) | How to parse the output |

Parser Configuration

| Field | Description | |-------|-------------| | resultsPath | JSONPath to the findings array (e.g., results, data.findings) | | fieldMapping.id | Path to finding ID | | fieldMapping.severity | Path to severity level | | fieldMapping.title | Path to finding title | | fieldMapping.description | Path to description | | fieldMapping.filePath | Path to file path | | fieldMapping.startLine | Path to line number |

Enable your custom scanner:

MIMIR_SCANNERS=semgrep,my-scanner bun run --cwd /path/to/mimir-scan start

Development

Prerequisites

• Bun runtime

Setup

git clone https://github.com/amitvishw/mimir-scan.git
cd mimir-scan
bun install

Running Locally

# Development mode with watch
bun run dev

# Run directly
bun run start

Testing

# Unit tests
bun test

# Unit tests with watch
bun test --watch

# E2E tests (requires scanners installed)
./test-e2e.sh

# E2E tests with Docker
./test-e2e.sh --docker

Building Docker Image

docker build -t mimir-scan .

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feat/amazing-feature)
3. Use conventional commits (feat:, fix:, docs:, etc.)
4. Run tests (bun test)
5. Submit a pull request

Commit Convention

This project uses Conventional Commits:

• feat: New feature (triggers minor version bump)
• fix: Bug fix (triggers patch version bump)
• docs: Documentation only
• chore: Maintenance tasks
• test: Adding tests
• refactor: Code refactoring

License

ISC

Links

• GitHub Repository
• npm Package
• Report Issues

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: amitvishw
• Source: amitvishw/mimir-scan
• License: ISC

Install and usage instructions live in the source repository linked above.