@codetitan/mcp-server

CodeTitan MCP Server — exposes the full Detect → Fix → Prove → Learn pipeline as MCP tools for Claude Code, Cursor, and any MCP-compatible client.

Tools

| Tool | Description | |------|-------------| | codetitan_analyze | Run static analysis (1001+ rules) across security, performance, refactoring, testing, and documentation domains | | codetitan_fix | Apply an AI-generated fix — dry-run to preview, single for one provider, arena to compete providers and pick the winner | | codetitan_arena | Run TitanArena: multiple AIs generate fixes in isolated git worktrees, real tests + static analysis choose the winner | | codetitan_insights | Show production feedback metrics: fix success rate, failing categories, auto-generated rule proposals | | codetitan_test | Detect the test framework (Jest/Vitest/Mocha/pytest/Go/Cargo) and run tests |

Setup

Prerequisites

npm install -g @codetitan/mcp-server
# or run directly with npx
npx @codetitan/mcp-server

The server requires @noalia/code-titan-core to be installed either globally or in the project you are analysing.

Claude Code

Add to ~/.claude/settings.json (global) or .claude/settings.json (project):

{
  "mcpServers": {
    "codetitan": {
      "command": "npx",
      "args": ["@codetitan/mcp-server"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "codetitan": {
      "command": "codetitan-mcp"
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "codetitan": {
      "command": "npx",
      "args": ["@codetitan/mcp-server"]
    }
  }
}

Claude Desktop

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

{
  "mcpServers": {
    "codetitan": {
      "command": "npx",
      "args": ["-y", "@codetitan/mcp-server"]
    }
  }
}

Usage Examples

Once connected, ask your AI assistant:

Analyse /home/user/my-project for security vulnerabilities

→ calls codetitan_analyze with domain: "security-god"

Fix finding SEC-001 in src/auth.ts using arena mode

→ calls codetitan_fix with mode: "arena" — multiple AIs compete, tests validate the winner

Run a quick analysis of the current project (cap at 20 files)

→ calls codetitan_analyze with quick: true

Show me production feedback metrics for the last 30 days

→ calls codetitan_insights with timeRange: "30d"

Run the tests for src/utils.ts

→ calls codetitan_test with the file path

Building from source

cd packages/mcp-server
npm install
npm run build   # outputs to build/
npm test        # 21 tests

Architecture

The server runs over stdio (standard MCP transport). Each tool call dynamically require()s the relevant module from @noalia/code-titan-core, falling back to a workspace-local install under <projectPath>/node_modules/. This means the server itself stays lightweight and always uses the core version that matches the analysed project.