crukx-mcp

AI-powered code reliability engine for any MCP-compatible coding agent.

Crukx deploys a 7-agent swarm (Stresser, Hunter, Red Teamer, Auditor, Scout, Stabilizer, Synthesizer) to audit your code for reliability, security, and correctness — directly inside your AI coding tool.

Works with VS Code + GitHub Copilot, Kiro, Cursor, Claude Desktop, opencode, and any other MCP-compatible client.

Install

npm install -g crukx-mcp

Login

crukx login
# Opens crukx.dev in your browser — sign in and return to terminal

Add to your editor

crukx install
# Auto-detects VS Code, Cursor, Claude Desktop, Kiro, opencode
# and configures them all in one step

Restart your editor. Done — your AI agent can now use Crukx tools.

Manual setup (if needed)

VS Code + GitHub Copilot

Add to .vscode/mcp.json in your workspace (or ~/.vscode/mcp.json globally):

{
  "servers": {
    "crukx": {
      "type": "stdio",
      "command": "crukx",
      "args": ["mcp"]
    }
  }
}

Then in VS Code: open the Chat panel → Agent mode → select crukx from the tools list.

Kiro (Amazon)

Add to ~/.kiro/settings/mcp.json:

{
  "mcpServers": {
    "crukx": {
      "command": "crukx",
      "args": ["mcp"],
      "env": {}
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "crukx": {
      "command": "crukx",
      "args": ["mcp"]
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "crukx": {
      "command": "crukx",
      "args": ["mcp"]
    }
  }
}

opencode

Add to ~/.config/opencode/config.json:

{
  "mcp": {
    "crukx": {
      "type": "local",
      "command": ["crukx", "mcp"],
      "enabled": true
    }
  }
}

Any other MCP client

Use stdio transport with:

• Command: crukx
• Args: ["mcp"]

Tools

Once connected, your AI agent can use these tools automatically:

swarm_audit

Full 7-agent reliability audit. Returns a score (0–100), grade (A–F), gate status, and per-finding fix suggestions.

"Audit this diff for me"
"Review the code I just wrote"
"Run a full reliability check on this PR"

security_scan

Targeted security analysis — injection attacks, hardcoded secrets, auth flaws, OWASP Top 10.

"Is this code secure?"
"Check for SQL injection in this query"
"Scan my auth handler for vulnerabilities"

reliability_score

Fast pass/fail gate — score + critical/high count only. No full findings list.

"Quick score before I commit"
"Is this safe to merge?"

suggest_fix

Concrete remediation for a specific issue or finding.

"How do I fix this SQL injection?"
"Give me a fix for the finding in swarm_audit"

report_fetch

Retrieve a previous audit report by ID.

"Show me report abc-123"
"Fetch the last audit"

crukx_health

Check gateway connectivity and auth status.

"Is Crukx connected?"
"Why are the tools not working?"

Example output

# Crukx Audit Report

**Score:** 74/100 (Grade C)
**Gate:** 🟠 REVIEW — high severity issues found
**Confidence:** 91%  |  **Agents:** Stresser, Hunter, Red Teamer, Auditor, Scout, Stabilizer, Synthesizer
**Report ID:** `a3f9b2c1-...`

## Severity Breakdown
| Severity   | Count |
|------------|-------|
| 🔴 Critical | 0    |
| 🟠 High     | 2    |
| 🟡 Medium   | 5    |
| 🔵 Low      | 3    |
| ⚪ Info     | 1    |

## 🟠 High Severity
- **SQL_INJECTION** — `src/db.ts:42`: User input concatenated into query
  > Fix: Use parameterized queries — `db.query('SELECT * FROM users WHERE id = $1', [userId])`

Links

• crukx.dev — Dashboard and reports
• GitHub