Skip to content

Tools reference＃

The kluster.ai Code MCP server provides review tools for checking AI-generated code quality and security. These tools enable real-time code reviews directly within your IDE through MCP integration.

It includes:

  • kluster_code_review_auto: Automatically reviews code quality and detects bugs, including logic errors, security issues, and performance problems. Triggers automatically when code is generated or modified. Best for real-time reviews during active coding sessions, analyzing changes in context of the full conversation and related files.
  • kluster_dependency_validator: Validates the security and compliance of packages and dependencies. Triggers automatically before package installations or when package files are updated. Best for preventing vulnerable or non-compliant third-party libraries from entering your codebase before installation.
  • kluster_code_review_manual: Manually reviews specific files when explicitly requested by the user (e.g., "review this file", "check for bugs"). Best for auditing existing code, reviewing specific modules, or getting fix recommendations for individual files.

This page documents the parameters and response formats you'll see when using these tools in Cursor, Claude Code, or any MCP-compatible client.

Parameters＃

These tools analyze AI-generated code and dependencies to detect bugs, security vulnerabilities, and other quality issues.

Parameters

code_diff string required

Unified diff format showing the actual changes (additions and subtractions) made to files. Use standard diff format with --- filename and +++ filename headers, followed by @@ line numbers @@, and + for additions, - for deletions. In MCP environments, this is often auto-extracted from IDE history.

user_requests string required

A chronological sequence of all user messages and requests in this conversation thread, with the current request (that triggered this assistant turn) clearly marked. Format: Previous requests as numbered list, then current request marked with >>> CURRENT REQUEST: [request text]. In MCP environments, this is often auto-extracted from conversation history.

modified_files_path string required

Full absolute paths of modified files separated by ;.

chat_id string optional

Session identifier returned by previous tool calls. Used to maintain context across multiple review requests.

The manual review tool is triggered only when explicitly requested by the user.

Parameters

user_requests string required

Chronological sequence of user messages with current request marked as >>> CURRENT REQUEST:. Unlike auto review, this parameter is NOT auto-extracted in MCP environments and must be explicitly provided.

modified_file_path string required

Full absolute path of the single file to review. This tool can only check one file per call.

need_fixes boolean required

Set to true if user requested fixes, false if only requesting issue detection.

chat_id string optional

Session identifier returned by previous tool calls. Used to maintain context across multiple review requests.

Response fields＃

All code review tools return the same response structure:

  • isCodeCorrect: Boolean indicating if the code has issues.
  • explanation: Summary of all issues found.
  • issues: Array of detected problems with:
  • type: Issue category (intent, semantic, knowledge, performance, quality, logical, security).
  • severity: Impact level (critical, high, medium, low).
  • priority: Execution priority (P0-P5).
  • description: Brief issue summary.
  • explanation: Detailed issue explanation.
  • actions: Recommended fixes.
  • priority_instructions: Execution rules for addressing issues.
  • agent_todo_list: Prioritized list of fixes to apply.
  • chat_id: Session identifier for maintaining context across requests.

Example response＃

{
    "isCodeCorrect": false,
    "explanation": "Found 1 issue. 1 critical issue needs immediate attention.\n\nTODO:\n1. [CRITICAL] The implementation introduces a critical SQL injection vulnerability.",
    "issues": [
        {
            "type": "intent",
            "severity": "critical",
            "priority": "P0",
            "description": "The implementation introduces a critical SQL injection vulnerability, which is an unacceptable security risk.",
            "explanation": "The code constructs an SQL query using string concatenation with user input, which is the classic pattern for SQL injection. A function designed for database interaction should use parameterized queries.",
            "actions": "Use parameterized queries or prepared statements to safely handle user input. For example: db.query('SELECT * FROM users WHERE id = ?', [userId])"
        }
    ],
    "priority_instructions": "**PRIORITY EXECUTION RULES:**\n1. **INTENT Critical/High (P0-P1) get special priority**\n2. **All other issues sorted by severity** - Critical (P2) > High (P3) > Medium (P4) > Low (P5)\n3. **Never let lower priority issues override higher priority changes**",
    "agent_todo_list": [
        "**EXECUTE IN THIS EXACT ORDER:**",
        "",
        "**Priority P0 - INTENT CRITICAL (HIGHEST PRIORITY):**",
        "P0.1: The implementation introduces a critical SQL injection vulnerability - Use parameterized queries or prepared statements."
    ],
    "chat_id": "i8ct930591"
}

Priority system＃

Code review assigns priority levels to detected issues, helping you focus on the most critical problems first. The system automatically prioritizes based on issue type and severity.

  • P0-P1: Intent issues (highest priority) - code doesn't match request.
  • P2: Critical severity - must fix immediately.
  • P3: High severity - should fix soon.
  • P4: Medium severity - nice to fix.
  • P5: Low severity - optional improvements.

Next steps＃