evidence-first code intelligence

Ask your codebase why.

The answer to one real engineering question is usually scattered across a dozen files. Meeba Brain traces it — and shows exactly which files back up every claim it makes.

Meeba Brain is a read-only GitHub tool that grades every AI answer about your code as Verified, Inferred, or Unknown, so you know what to trust before you act on it.

Try it on your own bug → See how it traces evidence
Read-only GitHub access Free while in early access Works inside Cursor & Claude Code
meeba.ai/brain
You: why does /checkout retry silently on payment failure?
verified The retry happens in payment_worker.py:88 — a bare except swallows the gateway timeout before the retry queue re-fires.
inferred This likely traces back to a change in #a41f9c2, where the timeout handler was widened to catch all exceptions.
unknown Whether this was intentional for the Stripe migration isn't in the repo — worth asking whoever wrote that commit.
3 files tracedgraph built from AST, not text search

Most AI tools guess with confidence.
This one shows its work.

Meeba Brain doesn't summarize your code from nearby text. It parses the actual structure — callers, callees, routes, dependencies — into a graph, then grades every claim against it before the answer ships.

01 · parse

Real AST parsing across your repo builds callers, callees, routes and dependencies — not a proximity guess from embeddings.

→

02 · graph

That structure becomes a queryable code graph, so an answer about one function can trace every place it's actually used.

→

03 · grade

Every claim ships as Verified, Inferred, or Unknown. Unverified references get re-checked before you see them, not after you catch the mistake.

cursor · chat
terminal
// MCP: meeba-brain connected
you: which services call refundOrder()?

meeba: 3 verified callers —
  billing-service, admin-panel,
  and a cron job in jobs/reconcile.ts

Works where you already work.

The same grounded graph, right inside the tools on your machine right now — no new tab, no new habit to build.

CursorClaude CodeMCP serverGitHub repo connect

Try it without handing over write access.

Meeba Brain needs to read your code to build the graph — it never needs permission to change it. Connect a repo the same way you'd share a link.

What GitHub access actually grants

The permission scope is read-only, on purpose. Nothing below is possible, even if asked.

✓
Read repository contentsNeeded to parse the AST and build the code graph.
✕
No write or push accessMeeba Brain cannot modify or commit to any file.
✕
No delete accessIt cannot remove branches, files, or the repository itself.
lowest-friction start

Have a public repo?

Skip the access conversation entirely — point Meeba Brain at any public GitHub repo and try a real question before connecting anything private.

Try on a public repo →

"I tested it on a real PDF parsing bug. It understood the underlying error context and gave a more accurate fix than the AI tools I normally reach for."

Akshat Tamrakar · Full Stack Developer, tested on a live production bug

Questions worth answering upfront.

The things people ask before connecting a repo.

What is Meeba Brain?

A read-only GitHub tool that traces AI-generated answers about your codebase back to real files and grades every claim as Verified, Inferred, or Unknown, instead of guessing from nearby text.

Does Meeba Brain need write access to my GitHub repo?

No. It only requests read-only access to repository contents to build its code graph. It cannot write, push, or delete branches, files, or the repository, even if asked.

How does Meeba Brain grade its answers?

Every claim is labeled Verified, Inferred, or Unknown based on whether it's directly backed by the code graph, likely but not certain, or not determinable from the repo at all.

Is Meeba Brain free to use?

Yes, it's free during early access. No credit card or phone number required to sign up.

Can I use Meeba Brain inside Cursor or Claude Code?

Yes. It runs as an MCP server, so the same evidence-graded code graph is available inside Cursor and Claude Code, not just in the web chat.

Can I try it without connecting a private repo?

Yes. Point Meeba Brain at any public GitHub repository and ask a real question before connecting or granting access to a private one.

How is Meeba Brain different from asking an AI coding assistant directly?

Most AI coding assistants answer from text proximity or embeddings, which can sound confident while being wrong. Meeba Brain builds an actual AST-based code graph of your repository, so answers trace to real callers, callees, and files — and every claim ships graded as Verified, Inferred, or Unknown.

Do I need to install anything to use it with Cursor or Claude Code?

Meeba Brain runs as an MCP server that connects to Cursor and Claude Code, so once it's connected you can query the same code graph from inside your editor without switching tabs.

Who is Meeba Brain for?

Engineers debugging a real issue in their own codebase who want an answer backed by evidence from the actual repository, rather than a plausible-sounding guess.

Bring one real bug.
Four minutes is enough.

Don't evaluate it in the abstract. Connect a public repo, paste the bug you're stuck on right now, and see whether the evidence holds up.

Free during early accessNo phone number needed

no credit card · read-only GitHub access

Create free account → Already have an account? Sign in

Signup takes under a minute — just email and a password, or continue with Google or GitHub.