mozg.whyexploreguideconnectSign in

Model Context Protocol · Claude Code · Codex · Cursor

Teach it once.
Every agent knows.

Drop in screenshots and files. mozg turns them into a searchable brain your coding agents read over MCP — and tells you how much it still doesn't know.

Build a brainBrowse public brains
claude code
$ claude mcp add --transport http mozg https://mozg.sh/mcp
✓ connected · 3 brains available
> build the pricing page — make it match our design system
brain_search(brain: "design", query: "card, spacing, price type")
→ 6 notes · 128 ms
Cards: 1px #E4E4E7 border, no shadow at rest, 8px radius, 24px pad.
Section gap is 32px — 24px is the inner scale, never between sections.
Price is 40/44 with tabular-nums. Primary CTA fills only on mobile.
Reads inClaude CodeCodexCursorClineKimi CLIQwen CodeVS Codeall of them →

What people put in one

The things you explain twice a week.

Design systems

Our components, not Tailwind's

Exact spacing, the states you actually ship, and the three rules everyone breaks.

Backend & APIs

The integration as it really runs

Your retries, your idempotency keys, the webhook order — not the vendor's happy path.

Game development

Engine conventions and math

How a mechanic is wired, what the pipeline expects, which numbers are load-bearing.

Product & process

What nobody wrote down

Naming, review rules, the deploy sequence — the folklore a new agent never gets.

What nobody else does

Every brain sits an exam.

You write what the brain is for. mozg turns that into control questions and runs them after every upload. So you get a number you can trust — and a list of exactly which material is missing, instead of guessing why your agent still gives bad answers.

Brain · Design system

32 checks · 4 categories
84%
Colour, borders and elevation12 / 12
Type scale and spacing8 / 8
Motion and transitionsmissing · no screen recordings yet3 / 7
Empty and error statesmissing · no source covers this0 / 5

One brain, every agent

Claude learns it on Monday, Codex knows it on Tuesday. Built-in agent memory stays locked to one vendor; a brain does not.

Agents write back

An agent that works out a convention saves it. You approve it in a review queue, so the brain gets sharper instead of noisier.

Yours to take

Export any brain as CLAUDE.md, a Claude Skill or AGENTS.md. It keeps working if you stop paying us.


Start here

Four pages, then you're running.

01

Why this exists

What a brain is, and why agent memory built into one vendor isn't the same thing.

02

Build a good brain

What to feed it, how to write a goal that becomes a real exam, and what a low score means.

03

Connect your agent

Copy-paste setup for Claude Code, Codex, Cursor, Kimi, DeepSeek, GLM and Qwen.

04

Or take one that exists

A catalogue of brains other people built, by field, free and paid.

In the catalogue now

all brains →

Stop explaining the same thing.

The free plan holds one brain, fifty sources and three hundred agent calls a month — enough to find out whether this works for you.

Build a brain