Check

Name: Check
Author: golproductions

@golproductions

关于 Check

Anti-hallucination firewall for AI coding agents. Validates commands before execution, imports against your package.json, function calls against your source files. Sub-100ms. $0.0068 AUD per check, 120 free on sign-up.

#command-validation #ai-agents #security #preflight #firewall #devtools #mcp

基本信息

分类

其他

传输方式

stdio

发布者

golproductions

提交者

Mr Traumatik

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "Check": {
      "command": "npx",
      "args": [
        "@golproductions/check-mcp"
      ],
      "env": {
        "GOL_CLIENT_ID": "your_key"
      }
    }
  }
}

工具

Validates a command, returns `RUNNABLE` or `INVALID`

Validates then executes, blocking invalid commands before they reach the shell

概览

What is Check?

Check is an MCP server that acts as an anti-hallucination firewall for AI agents. It validates commands, imports, and function calls before execution, returning a verdict of RUNNABLE or INVALID in under 100 milliseconds. It contains no AI of its own.

How to use Check?

Install by adding a configuration entry to your MCP client's mcpServers, specifying the command npx -y @golproductions/check-mcp and setting the environment variable GOL_CLIENT_ID to your API key. After setup, you can invoke the Check tool to validate a command or the CheckAndExecute tool to validate and then run it, blocking invalid commands before they reach the shell.

Key features of Check

Validates commands, imports, and function calls prior to execution.
Returns verdicts in under 100 milliseconds.
Free tier of 120 checks on signup.
No AI inside – purely deterministic validation.
Works with any MCP-compatible client.
Blocks invalid commands before they reach the shell.

Use cases of Check

Preventing AI agents from executing hallucinated or unsafe shell commands.
Securing development workflows that rely on AI-generated code with function calls.
Adding a validation layer before running imports or external commands in automated pipelines.
Auditing agent-generated actions in real time without slowing down the process.