Skip to content

Latest commit

 

History

8 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mcp-change-review

Before trusting a new MCP setup, review what changed.

Use mcp-change-review when installing, updating, or sharing MCP configuration for Claude Code or Codex. It shows which MCP servers changed, what local resources they can access, and obvious risks such as secrets, shell commands, and sensitive paths.

Node 20 plus Claude Code supported Codex supported MIT license

English | 简体中文

What it does

  • Discover MCP configuration used by Claude Code and Codex.
  • Create a local baseline and report added, removed, or changed MCP servers.
  • Flag secrets, sensitive paths, command execution, and Docker latest.
  • Print results in the terminal or export Markdown and JSON reports.

Quick start

curl -fsSL https://raw.githubusercontent.com/mctang24/mcp-change-review/main/install.sh | sh

Requires Node.js 20+ and npm.

Commands

  • mcpcr diff - Compare the current MCP config with the baseline; the first run creates .mcpcr-baseline.json.
  • mcpcr accept - Mark the reviewed state as trusted and save it as the new baseline.
  • mcpcr export md - Generate a Markdown report.
  • mcpcr export json - Generate a JSON report.
  • mcpcr diff --fail-on high - Exit non-zero in automation when high-risk changes are found.
  • mcpcr list - List discovered MCP servers.
  • mcpcr status - Check whether the current directory has a baseline.

Example

An MCP server gives an AI agent additional capabilities, such as reading local files or calling external services.

In this Codex example, filesystem is listed as a new MCP server, while github is flagged for exposing a credential-style environment variable name.

Example mcpcr diff output for Codex MCP changes

In this example, github is flagged because the configuration exposes a credential-style environment variable name, not because the GitHub MCP server is inherently unsafe.

Safety model

mcp-change-review reads local MCP configuration and reports risks using fixed rules.

  • It never stores secret values.
  • It never modifies Claude Code or Codex configuration.
  • It records env/header names only.
  • It does not proxy, block, or intercept MCP tool calls.
  • It uses fixed checks instead of LLM-based risk scoring.

Supported clients

Client Scope
Claude Code local, project, and user-level MCP configuration
Codex user-level and trusted project MCP configuration

Validated against Claude Code 2.1.195 and Codex CLI 0.142.3.

License

MIT

About

Review MCP config changes for Claude Code and Codex before new tools, tokens, or filesystem access reach your AI agent.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages