devswha/insane-review-mcp

Run insane-review (GPT Pro web review) from any harness — pipx CLI + MCP server over the pinned upstream engine (fivetaku/insane-review).

About insane-review-mcp

Run insane-review from any harness — not just Claude Code.

GPT Pro (the Pro reasoning tier of the current flagship, currently GPT-5.6 Sol) has no API. insane-review reaches it anyway by driving your logged-in ChatGPT web session over CDP: it repomix-packs the relevant code, sends it to Pro, and harvests the review at zero API cost. The upstream project ships that as a Claude Code plugin.

This repo makes the same engine usable from any harness — via a pipx CLI and an MCP server — without forking the engine.

Not a fork. The engine (packandask.py, MIT, by @fivetaku) is pinned by commit SHA and fetched on first use, never copied into this repo. One canonical engine, no duplicated maintenance. This repo only adds the cross-harness distribution layer (packaging + MCP + adapters).

upstream fivetaku/insane-review this repo --------- Engine (packandask.py) canonical home pinned + fetched (not copied) Claude Code plugin ✅ use upstream's pipx CLI on $PATH — ✅ insane-review MCP server (any MCP host) — ✅ insane-review-mcp Harness adapters (Codex, generic, …) — ✅ adapters/

If you only use Claude Code, install upstream directly — you don't need this.

On first use the pinned engine is fetched into ~/.cache/insane-review-mcp/. Refresh it with insane-review --update-engine; inspect the pin with insane-review --engine-info.

One-time browser bridge (required for a live review)

Pro is web-only, so you need a logged-in browser on a debug port:

From the project README.

Maintaining this server?

Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:

[![On MCP Radar](https://mcp.liqiwa.com/badge.svg)](https://mcp.liqiwa.com/s/devswha--insane-review-mcp.html)

Related MCP servers