HOW MOTIFS WORKS

From first idea
to what’s next.

Three skills run in the coding agent you already use. Each reads what the last one wrote, asks you only what it can’t find, and pulls researched plays from Motifs on the way.

Start with one play Questions

Made for your coding agentCodexClaude Code

FROM FIRST IDEA TO WHAT’S NEXT

Marketing belongs in
your product’s DNA.

It’s more than what you say at launch. It’s what you choose to build,
how it feels to use, and why someone comes back.

Claude, help me plan a habit app for people learning a language

motifs MCP connectedTry it in Claude Code

01 · Plan

Start from scratch

“Tell me what you want to build, who you imagine using it, and what you want it to help them accomplish.”

Then one to three at a time

Who are you building it for first?
How do they handle this today?

Planning Head · writesProject MD

01 · Plan

Start from your repo

README and landing copy, done
Pricing and billing code, done
Onboarding screens, in progress

Drafts what the code says, then asks only about the gaps. Motifs never sees the code.

Planning Head · writesProject MD

02 · Build

Build from the plan

Reads Project MD first, done
Feature: review each item just before it's forgotten, done
Research suggestion, with its limits, in progress

Execution Head · writesCore Features MD

03 · Go to market

Pick the play, then test it

Compares 3 reference plays on

Outcome evidence, done
Whether it transfers here, done
What the build supports, in progress
Then one small first test

GTM Head · writesGTM MD

Project MD2 agreed2 provisional1 open

SectionDecisionBorrowed from

Product descriptionAgreedA daily speaking habit for adults learning a languageYour answer—

Problem and desired progressAgreedLessons never turn into speaking. They want to hold a real conversationYour answer—

Sample use caseProvisionalTen minutes of speaking on the commute, the month before a tripExample, not observed—

OfferProvisionalDaily practice free; paid unlocks feedback on speakingYour answerLock everything, or give the core awaySourced · 7 sources

It factorOpenWhy would someone choose this over their current solution?Question to resolve—

Example output. Pick either plan card to switch. The product and its answers are illustrative; every play named is a real record.

THREE HEADS, ONE THREAD

Each step writes down
what the next one reads.

Nothing lives in a chat that scrolls away. Every decision lands in a markdown file in your project,
so the agent that builds knows what the agent that planned agreed with you.

01 · Plan
Planning Head
ReadsYour idea, or your repoPlays you pick, as inspiration
AsksOne to three questions at a time
WritesProject MD
- 1Product description
- 2Problem and desired progress
- 3Sample use case
- 4Offer
- 5It factor
02 · Build
Execution Head
ReadsProject MDPlays for the feature
AsksOnly what would change the answer
WritesCore Features MD
- 1Feature
- 2What it does
- 3User benefit
- 4Live, unreleased or partly built
03 · Go to market
GTM Head
ReadsProject MDCore Features MDOne or more reference plays
AsksGoal, budget and audience, in small batches
WritesGTM MD
- 1Additional GTM context
- 2Play comparison and selection
- 3Application to this company
- 4First test

THE PLANNING HEAD

It asks. You decide.

Start with one rough sentence, or point your agent at the repo. The Planning Head turns either into Project MD: five sections, and each decision marked by who made it.

From scratch, it asks one to three questions at a time: the ones that would change the direction, never the whole checklist.
From your repo, it drafts what the code already says, marks it Provisional, and asks only about the gaps.
Plays you pick come in as inspiration. They never override what you want to build.

It factorOpen

What makes the experience noticeably different?

Not in the repo

Planning HeadWhen someone opens it, what do they get that their current flashcards don’t?

YouEach word comes back the day they’re about to forget it.

It factorAgreed

Reviews timed to each learner’s memory, not a fixed schedule

Your answer Memory-model review queue

Agreed: You said it. Agreement is not validation: what sits under it is still an assumption.
Provisional: Written down as an assumption, with what is unresolved and what would settle it.
Open: A question worth answering before anything is built on it.

THE EXECUTION HEAD

Every feature arrives with its limits.

The Execution Head reads Project MD before anything else. If the direction isn’t settled, it goes back and plans with you first, keeping whatever you already agreed.

A feature comes with the smallest useful version and how you’ll know it helped.
A research-based suggestion doesn’t have to add anything. Sometimes the answer is to remove a step.
It writes Core Features MD: what is live, unreleased or half-built, so nothing gets marketed before it exists.

Execution Head · FeatureExample

Review each item just before it's forgotten

Project fit: Problem and desired progress: words are forgotten between sessions
Play basis: +16.5% recallMemory-model review queue · 5 sources · strong

Kept

A per-learner memory model decides when each word returns

Changed

Words, not course items. Reviews sit inside the daily session

Left out

Per-item difficulty weights

What the evidence does not establish

One teacher, one school, one course; students were not told the details of the manipulation. Outcome is exam recall, not engagement or voluntary return. Read as the authors' manuscript hosted on Mozer's site; the published version was not compared.

THE GTM HEAD

The biggest number doesn’t win by default.

Hand it one play or several. The GTM Head weighs each against the same goal: the outcome evidence, whether it transfers to you, and what your build supports. Then it picks, and says how sure it is.

It asks for goal, budget and audience in small batches, and records the answers in GTM MD so nobody asks twice.
“Not feasible now” is not “unlikely to work”. It says which.
The first test is chosen for what it teaches, not as a prediction that it wins.

GTM MDPlay comparison and selection

Price per countryCOMPANY-REPORTED+50% priceNeeds enough users in each country to split into test groupsNot feasible now

Store page per keywordCOMPANY-REPORTED−25% CPANeeds paid search campaigns, and none are running yetBlocked by a prerequisite

Purchase button wordingCOMPANY-REPORTED+15%The paywall is live, and labels can be split without a releaseFirst test

First testExample

Two button labels, two groups

Success measure: Trial starts ÷ paywall views, per group
Guardrail, from the play: A generic “Continue” on the step that starts a charge does not say a subscription will begin; Google Play cites a final “continue” tapped without meaning to as a violation.

ContinueAdjustStop

Thresholds are proposed decision points, not benchmarks.

WHAT YOUR AGENT PULLS

Plays, with the fine print attached.

Each head reaches into the same library. A finding never travels without its limits,
and every row says how strong its evidence is.

Your repoRead by your agent, on your machine

Motifs MCPAnswers with records

Motifs gets the search, never your code.

motifs · tool calls

search_plays funnel_stage: retention · min_evidence: controlled_experiment→ summaries, strongest evidence first

get_play memory-model-review-queue→ the whole record: mechanism, sources, evidence, when to avoid it

STRONGLindsey et al. · 2014

Finding

Among 179 eighth graders in one Spanish course, items reviewed on a per-student memory model were recalled 16.5% better on an exam 28 days after the semester than items given massed review, and 10.0% better than a fixed one-week spaced schedule — with the number of reviews held equal.

Limits

STRONG
Peer-reviewed work, or a controlled experiment, that tests this mechanism directly.
SUPPORTING
Credible research on an adjacent question, or a result that is correlational rather than causal.
COMPANY-REPORTED
The company’s own published numbers. Not independently verified, and they chose what to publish.
UNTESTED
Untested: a mechanism we find plausible, or one person’s account of what happened to them. Nothing controlled behind it.

WRITTEN INTO EACH SKILL

What it won’t do.

The rules the three heads keep, whoever asks.

Every head
It won’t Invent evidence
No made-up studies, statistics or customer evidence. A number keeps its population and its caveat.
Planning Head
It won’t Mark a guess as agreed
What it infers from your repo stays Provisional until you confirm it.
Execution Head
It won’t Rewrite your direction to fit a play
Plays adapt to your product. Project MD changes only with your agreement.
GTM Head
It won’t Let the famous company win
The biggest result or best-known name doesn’t rank first by default.
GTM Head
It won’t Market what isn’t built
Unreleased or partial features are never promised as available.
GTM Head
It won’t Publish, contact or spend
A plan authorises nothing. It acts only on what you explicitly ask.

BEFORE YOU START

Questions, answered.

Which coding agents does it work with?

Any agent that speaks MCP. The examples on this page use Claude Code, Codex and Cursor; the workflow is the same in each, and only the way you connect differs.

Do I have to start from scratch?

No. Point your agent at an existing repo and the Planning Head drafts Project MD from what the code already says: the README, pricing, onboarding. Everything it infers starts Provisional, and it asks only about what the code can’t tell it.

Does Motifs see my code?

No. Your agent reads your repo on your machine. What reaches Motifs is a search for plays, and what comes back is the records.

Where do Project MD, Core Features MD and GTM MD live?

In your project, as plain markdown your agent writes, somewhere you agree on. They’re yours to read, edit and commit. Each head reads the ones before it, so a question you’ve answered once isn’t asked again.

Can I skip straight to building or launching?

You can ask, but the Execution Head won’t recommend features without a settled Project MD. It goes back to planning, keeps what you’ve already agreed, and asks only what’s missing. The GTM Head needs at least one reference play and won’t substitute generic marketing advice for one.

How much should I trust a play?

As far as its evidence goes, and each record says how far that is. Every evidence row carries its tier and what it does not establish, side by side. All plays today are Sourced: built from published experiments and company reports, not yet checked on a dated day.

Will it change my product on its own?

No. A feature recommendation is not permission to build it, and a GTM plan is not permission to publish, contact anyone or spend money. It implements only what you ask, and edits Project MD only with your agreement.

What does it cost?

Every play is free to read, and five one-click copies a month are free. Searching through the MCP needs any signed-in account; whole records through the MCP are part of all-access. See pricing.

Is all of this ready today?

The library is live and free to read. The MCP and the three heads are in early preview: founding members get each piece as it ships. The product on this page is an illustration; every play it names is a real record.

Start with
one play.

Copy it into the agent you already use. No account needed to browse.

Start with one play Explore the plays

It asks. You decide.

Start with one rough sentence, or point your agent at the repo. The Planning Head turns either into Project MD: five sections, and each decision marked by who made it.

From scratch, it asks one to three questions at a time: the ones that would change the direction, never the whole checklist.

From your repo, it drafts what the code already says, marks it Provisional, and asks only about the gaps.

Plays you pick come in as inspiration. They never override what you want to build.

Every feature arrives with its limits.

The Execution Head reads Project MD before anything else. If the direction isn’t settled, it goes back and plans with you first, keeping whatever you already agreed.

A feature comes with the smallest useful version and how you’ll know it helped.

A research-based suggestion doesn’t have to add anything. Sometimes the answer is to remove a step.

It writes Core Features MD: what is live, unreleased or half-built, so nothing gets marketed before it exists.

The biggest number doesn’t win by default.

Hand it one play or several. The GTM Head weighs each against the same goal: the outcome evidence, whether it transfers to you, and what your build supports. Then it picks, and says how sure it is.

It asks for goal, budget and audience in small batches, and records the answers in GTM MD so nobody asks twice.

“Not feasible now” is not “unlikely to work”. It says which.

The first test is chosen for what it teaches, not as a prediction that it wins.

From first ideato what’s next.

Marketing belongs inyour product’s DNA.

Start from scratchFrom scratch

Start from your repoYour repo

Build from the planBuild

Pick the play, then test itGo to market

Each step writes downwhat the next one reads.

Planning Head

Execution Head

GTM Head

It asks. You decide.

Every feature arrives with its limits.

Review each item just before it's forgotten

The biggest number doesn’t win by default.

Two button labels, two groups

Plays, with the fine print attached.

What it won’t do.

It won’t Invent evidence

It won’t Mark a guess as agreed

It won’t Rewrite your direction to fit a play

It won’t Let the famous company win

It won’t Market what isn’t built

It won’t Publish, contact or spend

Questions, answered.

Start withone play.

From first ideato what’s next.

Marketing belongs inyour product’s DNA.

Start from scratchFrom scratch

Start from your repoYour repo

Build from the planBuild

Pick the play, then test itGo to market

Each step writes downwhat the next one reads.

Planning Head

Execution Head

GTM Head

It asks. You decide.

Every feature arrives with its limits.

Review each item just before it's forgotten

The biggest number doesn’t win by default.

Two button labels, two groups

Plays, with the fine print attached.

What it won’t do.

It won’t Invent evidence

It won’t Mark a guess as agreed

It won’t Rewrite your direction to fit a play

It won’t Let the famous company win

It won’t Market what isn’t built

It won’t Publish, contact or spend

Questions, answered.

Start withone play.

From first idea
to what’s next.

Marketing belongs in
your product’s DNA.

Start from scratch

Start from your repo

Build from the plan

Pick the play, then test it

Each step writes down
what the next one reads.

Start with
one play.

From first idea
to what’s next.

Marketing belongs in
your product’s DNA.

Start from scratch

Start from your repo

Build from the plan

Pick the play, then test it

Each step writes down
what the next one reads.

Start with
one play.