Urgent.News

What's breaking now, across thousands of outlets.

AI

Get a second-opinion code review from Claude Code over MCP

Your coding agent is good at reviewing code. It is less good at reviewing its own code: it reads the change with the same prompt, the same context and the same assumptions that produced it. For risky changes (authentication, payments, data deletion) a reviewer from outside that loop is worth having. This post shows how to ask an independent AI agent for a review from inside Claude Code, using the…

In a world where AI can now assist with code reviews, a new method has emerged to ensure accuracy and objectivity. The Model Context Protocol (MCP) allows for independent AI agents to review code outside of the traditional loop, providing a second opinion on potentially risky changes.

To begin, add a remote MCP server using the claude mcp add command, specifying the transport protocol, scope, and the URL of the MCP server. In this example, the Speedbot.dev MCP server is used as a read-only planning operation. Once the connection is confirmed, request a read-only plan from Claude Code, specifying the desired review details such as the independent code review of changes in a specific pull request.

The reviewer will return a prioritized report with file and line references, the checks performed, and any unverified aspects.

Craft a brief for the reviewer, including the repository, pull request, expected behavior, and test commands. Clearly state what constitutes a satisfactory outcome and what should not be shared, such as secrets or private data. Before accepting the review, run the provided test command and verify high-impact findings at the referenced lines. If a finding lacks a reproducible check, treat it as a hypothesis rather than a confirmed defect.

After receiving the second review, compare it with the original review and the initial one. Keep only the findings that have been verified and note any checks that require additional access or context. The read-only plan serves as an effective way to test the MCP setup and ensure the reviewer receives the appropriate scope. Keep in mind that Speedbot.dev maintains the example MCP server used in this example.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at dev.to →

More in AI

I built a 50-year-old security model for LLM agents. It caught things I didn't expect.

Prompt injection is the problem nobody's actually solved. An agent reads an email or a web page, and hidden inside that data is an instruction that redirects it.

  • Kelvra developed a 50-year-old security model for LLM agents to prevent prompt injection attacks.
  • The model uses a policy file to designate sources, sinks, and permitted crossings between them.
  • Kelvra checks data flow before execution, identifying potential attacks without running the model.

More from Thursday 12 November →