A secure MCP (Model Context Protocol) server that allows AI assistants like Claude to execute TypeScript and JavaScript code through Deno.

• 🔒 Secure code execution using Deno's permission system
• 🚀 TypeScript and JavaScript support with full Deno runtime
• 🛡️ Configurable permissions set at server startup
• 🔌 MCP compliant for seamless AI assistant integration
• ⚡ Fast and lightweight with minimal dependencies

Install from JSR:

deno add jsr:@cong/mcp-deno

Or run directly:

deno run --allow-all jsr:@cong/mcp-deno

Add to your Claude Desktop configuration:

{
  "mcpServers": {
    "deno-executor": {
      "command": "deno",
      "args": [
        "run",
        "--allow-all",
        "jsr:@cong/mcp-deno"
      ]
    }
  }
}

Control what the executed code can access by modifying the startup arguments:

{
  "mcpServers": {
    "deno-executor": {
      "command": "deno", 
      "args": [
        "run",
        "--allow-net=example.com",
        "--allow-read=/tmp",
        "--allow-write=/tmp",
        "jsr:@cong/mcp-deno"
      ]
    }
  }
}

Available permission flags:

• --allow-net[=<allow-list>] - Allow network access
• --allow-read[=<allow-list>] - Allow file system read access
• --allow-write[=<allow-list>] - Allow file system write access
• --allow-run[=<allow-list>] - Allow running subprocesses
• --allow-env[=<allow-list>] - Allow environment access
• --allow-all - Allow all permissions (use with caution)

# Install dependencies
deno cache main.ts

# Run in development mode
deno task dev

# Run tests
deno task test

# Type check
deno check main.ts

For debugging and testing your MCP server, use the MCP Inspector:

# Install MCP Inspector
npm install -g @modelcontextprotocol/inspector

# Run the server with inspector
npx @modelcontextprotocol/inspector deno run --allow-all jsr:@cong/mcp-deno

# Or test locally
npx @modelcontextprotocol/inspector deno run --allow-all main.ts

This MCP server prioritizes security through:

• Subprocess isolation: Code runs in separate Deno processes
• Explicit permissions: Only granted permissions are available to executed code
• No persistent state: Each execution is independent
• Permission error handling: Clear feedback when permissions are insufficient

MIT

Contributions are welcome! Please open an issue or submit a pull request.