MegaLens.ai

From a recorded Cursor build

A check before AI mistakes reach your git history.

9 October 2026 note. Prices and models at the time of writing; see /pricing for current.

Reviewers from different AI companies check your change on their own. A judge marks a finding verified only when it can quote the code that proves it. It is not a linter: it looks for logic, security and behaviour bugs, not style or syntax.

No code to write. Works inside the Cursor chat you already use.

Try it free on your project

Free account. Your first review is free, no card needed.

Findings from one recorded build

Cursor reads findings like these and decides what to fix. The small line on each card is the finding exactly as it was recorded.

  • CriticalSecurity

    A repo's own git config could run code on your machine during normal commands.

    As recorded: Git config runs arbitrary code during normal commands

  • CriticalOperational

    A force-push could create duplicate PRs or endless retries.

    As recorded: Force-push breaks state tracking: duplicate PRs or retry loops

  • CriticalSecurity

    A huge repo could crash the tool by running git with no limits.

    As recorded: Git commands with no timeout or output limit. Huge repo = out of memory…

  • HighSecurity

    Another program on your computer could hijack the GitHub login.

    As recorded: The plan uses a fixed port (8765) for the OAuth callback. Any other program on the same machine can bind that port first…

From a recorded Cursor build. The full comparison is in the case study here.

From the recorded build

Severity table from the plan review of the recorded build, showing findings 6 to 15 with severity, category and the finding text.

Findings 6 to 15 of the plan review, from the recorded Claude Code build in the same comparison. There is no published recording of the Cursor run; its findings are in the full comparison linked above.

Three steps

1

Create a free account

Sign up free. No card needed, and your first review is on us.

2

Connect it to your coding tool

One command, pasted into the terminal on the computer where you use Cursor. This step needs that computer, not your phone.

Paste into your terminal
npx megalens-mcp setup

It finds Cursor, asks for your connection token, which you copy from your MegaLens settings, and writes the config file for you. Then restart Cursor and type run megalens_status in the chat. A reply means you are connected. To paste the config into .cursor/mcp.json yourself, use the exact snippet in the Cursor guide.

Claude Code and Codex each have their own setup guide: Claude Code · Codex

3

Ask for a review in plain English

Type either of these into your chat. Your coding tool takes it from there.

“Review this plan with MegaLens before we build”

“Run a MegaLens security audit on what we just built”

Example of what comes back in your chat

> Review this plan with MegaLens before we build

megalens_debate · done

Independent reviewers returned findings, including:

[Critical] Git config runs arbitrary code during normal commands

[Critical] Force-push breaks state tracking: duplicate PRs or retry loops

I checked these against the plan. Both hold up. Updating the plan before we build.

Shown as text, not a screenshot. The two findings are real, from the recorded build.

What happens next

  • The findings come back into your chat.
  • Cursor reads them and decides what to fix.
  • You stay in the same conversation.

You do not need to leave your IDE.

Findings like these, on your own project, in the chat you already use.

Try it free on your project

First review free. Then pick a plan, or pay as you go.

Questions

Do I need to know how to code?
No. You run one setup command, then ask for reviews in plain English inside the chat you already use. Your coding tool reads the findings and handles the fixes.
What is an MCP?
A small add-on that lets your coding tool talk to an outside service. The setup command in step 2 installs it for you, and after that you just type normal English.
How can I try it?
Create a free account and run your first review with no card and no API key. It covers up to 35,000 tokens of input, which is a plan or a few files. After that, pick a plan with reviews included, or pay as you go.
Will it send my code somewhere?
Yes. The code you ask about is sent to the AI models that review it. Details are in the privacy policy.
Does it replace Cursor?
No. It adds independent reviewers to it. Cursor still writes the code and still decides which findings to act on.
Is it a linter?
No. MegaLens ignores formatting and syntax and looks for logic, security and behaviour bugs. Keep your linter for style.
Does it check my pull requests by itself?
No. MegaLens runs when you or your AI tool ask for a review, so it can catch problems before a branch is pushed. It does not yet gate pull requests automatically.
What does it cost after the free review?

Two ways to pay.

A plan: Plus and Pro include a monthly allowance of reviews, with model usage included, from $19 a month. MegaLens selects up to four reviewing models. A review of up to 25,000 tokens of input counts as one; larger ones count one more for every further 25,000. Research and planning count double. The count is worked out from the size of your input as MegaLens estimates it (about 3.5 characters to a token).

Pay as you go: You keep a balance with MegaLens, and each review is charged from it. Your free review comes first. The current rate is on the pricing page.

Full details on the pricing page.

The full comparison lists every major finding from both builds. Read the full comparison