MegaLens checks what Claude Code writes before it reaches your git history.
Reviewers from different AI companies check your code. You call them from your terminal over MCP. A judge marks a finding verified only when it can quote the code that proves it. Setup is one config line, and you do not need to switch tabs.
How it works
Claude Code writes code. You invoke megalens_debate with a skill tag.
MegaLens selects up to four reviewers from different AI companies. They review your code independently.
A judge checks each finding against the code.
A finding is verified only when the judge can quote the code that proves it. Otherwise it is marked unverified or disputed. Findings point to the file and line when you send the code, and each result names the models that actually took part.
Claude Code receives the list.
It decides what to apply, reject, or investigate. Your IDE keeps the final say.
Setup
The quickest way is the setup wizard. It detects Claude Code and writes the config for you.
npx megalens-mcp setup
Detects Claude Code, asks for your token, and writes the config below automatically. Or paste it yourself:
Run this once. The --scope user flag makes MegaLens available in every project:
claude mcp add --transport http --scope user megalens \ "https://megalens.ai/api/mcp?ide=claude-code" \ --header "Authorization: Bearer ml_tok_your_token_here"
Or edit ~/.claude.json yourself:
{
"mcpServers": {
"megalens": {
"type": "http",
"url": "https://megalens.ai/api/mcp?ide=claude-code",
"headers": {
"Authorization": "Bearer ml_tok_your_token_here"
}
}
}
}Get your token at megalens.ai/app/settings/mcp. Restart Claude Code, then run megalens_status to confirm. Setup wizard source: github.com/megalens/mcp.
Using Claude.ai in the browser instead of Claude Code? Add MegaLens there as a custom connector. There is no token to copy: see the steps.
What MegaLens can catch in code from Claude Code
Security assumptions
Auth logic, token handling and CORS rules are where one model can be too trusting. A model from a different company can spot what the first one assumed was safe.
Framework misuse that causes bugs
Next.js, Rails, Django, Laravel. Each framework has conventions, and breaking them can cause real bugs. MegaLens looks for the bugs, not for style.
Contradictions across files
A type defined one way in the schema and used differently in the handler. If you send both files, the reviewers can catch a mismatch between them.
Overconfident rewrites
A model reviewing its own rewrite is not an independent check. A review from a different model family can catch regressions the author missed.
Honest limitations
Advisory only. MegaLens does not merge, push, or run tests.
Not a linter. It ignores formatting and syntax and looks for logic, security and behaviour bugs. Keep your linter for style.
It runs when you or your tool ask for a review. It does not yet gate pull requests automatically.
Review quality depends on how specific your request is. Name a skill when you can.
A review takes a few minutes. It is not for autocomplete.
Pricing
Free
$0
Your first review is free, with no card. MegaLens selects two reviewing models for it. After that, pay as you go or pick a plan.
Plus and Pro
From $19 a month
A monthly allowance of reviews with model usage included: 50 reviews on Plus, 200 on Pro. 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). Every plan and pay-as-you-go run the strongest reviewer panel, with reviewers from Anthropic, OpenAI and Google. Every review is verified the same way. Pro is the team plan: one account, one bill and one shared allowance. Cancel any time.
How model costs are paid
On a plan, model usage is part of your monthly allowance. Otherwise, pay as you go with prepaid credits. See all plans and rates
Check your code before it reaches git history.
Run your first MegaLens review inside Claude Code. It is free and needs no card.