Skip to main content
subagent CX Codex

mcp-reviewer — Claude Code subagent: MCP server review, how to install

mcp-reviewer reviews MCP server changes: tool schemas, safety gates on destructive calls, error handling and stdio correctness. Useful before you merge work on a server.

MIT ★ 1,198 updated 2026-10-23

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/agents/mcp-reviewer.md
Open on GitHub

What mcp-reviewer does

A reviewer subagent aimed at MCP servers. It reads the diff for tool names and how strict their schemas are, checks whether mutating and destructive calls sit behind a confirmation gate, and looks at timeouts, retries and error paths in the API client.

It runs on opus with four tools: Read, Grep, Glob, Bash. Stdio gets its own pass, since a server should log to stderr only. Anything written to stdout corrupts the message stream, and the review calls it out. The host config and runbook are checked against what actually runs.

You get findings sorted by priority, file references and commands to verify each one. Drop the file at .claude/agents/mcp-reviewer.md for Claude, or .codex/agents/mcp-reviewer.toml for Codex. MIT licensed, from hoangsonww/Claude-Code-Agent-Monitor.

What it helps with

  • check that destructive MCP tools sit behind a safety gate
  • catch stdout logging that breaks the stdio protocol
  • tighten tool schemas and rename confusing tools
  • add timeouts and retries to the server's API client
  • match the host config and runbook to the real launch commands

Example prompts for the agent

  • Run mcp-reviewer on the MCP server diff and list findings by priority
  • Review the tool schemas and safety gates in src/server.ts, with file references
  • Check stdio logging: is anything going to stdout?

How to install mcp-reviewer

Manually: put the files in place

.claude/agents/mcp-reviewer.md

For all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention

Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean

Questions about mcp-reviewer

How do I install mcp-reviewer in Claude Code?

Copy the subagent file into your project at .claude/agents/mcp-reviewer.md. That is the whole install; it then shows up under the name mcp-reviewer.

What does mcp-reviewer do?

It audits changes to an MCP server: tool names and schemas, gates on mutating and destructive operations, timeouts and retries in the API client, stderr-only logging, and whether the host config and runbook match reality. Output is prioritised findings with file references.

Which model and tools does mcp-reviewer use?

opus, with access to Read, Grep, Glob and Bash. That covers reading server code, searching the repo and running verification commands.

Source text

mcp-reviewer.md · MIT
---
name: mcp-reviewer
description: Review MCP server changes for tool safety, schema quality, and host integration correctness.
tools: Read, Grep, Glob, Bash
model: opus
---

You are an MCP-focused reviewer for this repository.

Focus on:
- Tool naming and schema strictness.
- Safety gate enforcement for mutating/destructive operations.
- API client timeout/retry/error handling.
- Stdio protocol safety (stderr-only logs).
- Host configuration and runbook documentation accuracy.

Output:
- Prioritized findings.
- File references.
- Verification commands to run.

© hoangsonww · MIT · hoangsonww/Claude-Code-Agent-Monitor

Similar subagents

All subagents for Agent setup: memory, context, orchestration →

From the same set155

The whole set →
frontend-reviewer The frontend-reviewer subagent reviews React UI changes for behavior regressions, state mismatches, and broken user flows. Use it after frontend edits. CX Codex ★ 1,198repo stars platform-status platform-status runs the CCAM diagnostics: hooks, config roots, dashboard version, MCP build state. Reports back without changing anything. CX Codex +1 ★ 1,198repo stars mcp-typescript Sets boundaries for TypeScript MCP servers: sends logs to stderr, keeps input schemas strict, and guards destructive actions. CX Codex +1 ★ 1,198repo stars run-operator run-operator — a subagent for operating Claude Code and Codex processes through CCAM, with a parameter check before each run, message, or stop action. CX Codex ★ 1,198repo stars backend-node backend-node is a Claude Code rule for Node.js backend files and hook ingestion logic. It keeps API responses compatible and enforces database transaction safety. CX Codex +1 ★ 1,198repo stars report report is a Claude Code command that creates CCAM executive, cost, reliability, or workflow reports. Specify a report type and data scope in the request. CX Codex +1 ★ 1,198repo stars release-guard release-guard is a Claude Code skill that checks a repository before merge or release, comparing documented commands with package.json and architecture notes with current code paths. CX Codex +1 ★ 1,198repo stars repo-onboarding repo-onboarding guides AI coding agents through codebase architecture, layer selection, and test command discovery before editing. CX Codex +1 ★ 1,198repo stars

Project sponsors

Companies that keep this analytics open