The brain your
agents are missing.

ContextsBase is context infrastructure for coding agents: features, business rules, data model, tests, design, served over MCP. Bring Claude, Cursor, or Copilot. We supply what they build from.

Connect your first agentSee how it works
  • Works with Claude Code, Cursor, Copilot, Windsurf
  • Bring your own agent
  • Free for one project
contextsbase · what you wrote
I-1Accounts
Open to agents
  1. 1F-1

    Sign up with email

    tests T-1, T-2

    PendingIn progressImplemented
  2. 2F-2

    Sign in with Google

    tests T-3

    PendingIn progressImplemented
  3. 3F-3

    Password reset

    tests T-7, T-8

    PendingIn progressImplemented
  4. 4F-4

    Two-factor auth

    tests T-9, T-10

    PendingIn progressImplemented
  5. 5F-5

    Delete account

    tests T-11

    PendingIn progressImplemented

5 features · 11 test cases · entities User, Session, ResetToken · published

MCP
your agent · claude code, cursor, copilot
  1. you ›Implement iteration I-1
  2. →Claimed F-1 Sign up with email
  3. ✓F-1 built · T-1, T-2 passed · recorded
  4. →Claimed F-2 Sign in with Google
  5. ✓F-2 built · T-3 passed · recorded
  6. →Claimed F-3 Password reset
  7. ✓F-3 built · T-7, T-8 passed · recorded
  8. →Claimed F-4 Two-factor auth
  9. ✓F-4 built · T-9, T-10 passed · recorded
  10. →Claimed F-5 Delete account
  11. ✓F-5 built · T-11 passed · recorded
  12. ●I-1 complete · 5 of 5 · 11 tests green

Five features you wrote. One instruction. The agent works the list.

Why agents ship the wrong thing.

// agent wrote

if (customer.tier === 'enterprise')

Customer has no tier field

They guess your data model

A pasted doc is not a schema.

// the usual e2e test

click('Reset password')

expect(toast).toBeVisible() ✓

  • link expires in 15 min?
  • link works only once?

Tests check buttons, not rules

The click works. The business rule is never asserted.

NotionSlack threadJiraYour head
copy, paste into chat, repeat

you › “Here’s the spec again…”

The spec lives somewhere else

So you paste it into every session.

You write the knowledge. Any agent builds from it.

  1. 01You

    Write the spec

    Rules and test cases.

    F-3 · Published
  2. 02You

    Queue it

    Add it to an iteration.

    I-1 · step 3
  3. 03Agent

    Pull it

    Gets spec, schema, tests.

    next_iteration_item
  4. 04Agent

    Build and test

    Code plus Playwright tests.

    T-7, T-8 passed
  5. 05Agent

    Record it

    What it built, against which spec.

    Implemented
  6. 06ContextsBase

    Watch it

    Spec edits flag the record.

    Outdated

One call. Everything the agent needs.

No pasting. No re-explaining.

  • SpecF-3 Password reset
  • EntitiesUser, Session
  • Flowreset-password
  • TestsT-7, T-8
  • Themetheme-v3
next_iteration_itemone MCP call
{
  feature:  "F-3 Password reset",
  spec:     [ 2 rules ],
  entities: [ User, Session ],
  flow:     "reset-password",
  tests:    [ T-7, T-8 ],
  theme:    "theme-v3"
}

Most tests prove the button works. Not that the rule holds.

ContextsBase test cases are written as business rules. Agents turn them into code that asserts the rule.

typical e2e test
test('reset password', async () => {
  await page.click('Reset password')
  await expect(page.getByText('Email sent')).toBeVisible()
})
Passes.Checks that a toast appeared.
contextsbase test case → generated test
T-7Expired link is rejected
Given
a reset link created 16 minutes ago
When
the user opens it
Then
they see “This link has expired” and no session is created
const link = await createResetLink({ ageMinutes: 16 })
await page.goto(link)
await expect(page.getByText('This link has expired')).toBeVisible()
expect(await sessions.count()).toBe(0)
Passes.Checks the business rule.

Design the look once. Agents build on-brand.

Pick colors, type, and radius. Preview live. Publish. Your agent pulls the tokens instead of guessing what “on brand” means.

contextsbase · theme

Primary

#3ECF8E

Font

Inter500 / 400

Radius

4px6px8px12px

Spacing base

8px

Live preview

Reset your password

We’ll email you a link. It expires in 15 minutes.

Send link
NewBetaBack to sign in
Published as theme-v3agents read the tokens

Click anything on your live page. Change it. Your agent ships it.

Words, spacing, colors. Saved as edits your agent applies to the real code.

contextsbase · page editor · yoursite.com/
yoursiteProductPricingStart
h1 · hero

Reset passwords without the support tickets.

One link. Fifteen minutes. Works once.

Start freeSee docs

Text

Reset passwords without the support tickets.

Style

Size 48 → 56
Weight 500

Saved changes

  • Hero headline texth1
  • Hero headline size56px
  • CTA label → “Start free”button
Save 3 changes
your agent

you › Apply the page edits

  • ✓ Pulled 3 saved edits for /
  • ✓ Hero.tsx · headline text and size
  • ✓ Hero.tsx · CTA label
  • ✓ Opened PR #42

Edits are read over MCP. The agent changes the real code, not a copy.

Order the work. Agents pull it one step at a time.

  1. 1

    You order. Put features in a list.

  2. 2

    Agents pull. Each agent takes the next one. Never the same one twice.

  3. 3

    Board updates. Done when the agent records it.

contextsbase · iterations
I-1Accounts
Open to agents
  1. 1F-1Sign up with email
    Done
  2. 2F-2Sign in
    Done
  3. 3F-3Password reset
    In progress
  4. 4F-4Two-factor auth
    Pending
  5. 5F-5Delete account
    Pending

Every team has a knowledge base for humans. Nobody has one for agents.

ContextsBase is the knowledge base agents read from.

Humans get

Agents get

A wiki with the rules

A pasted prompt

A data model diagram

A guess

Acceptance criteria

“Looks fine”

Everything above

With ContextsBase

Who uses it.

Solo founder

Re-explains the app every session.

Agent reads it from ContextsBase.

Platform team

Two agents build the same thing.

Each agent pulls a different step.

Agency

“That’s not what we agreed.”

The record shows the spec they approved.

QA lead

Test coverage lives in a spreadsheet.

Every test case shows its Playwright result.

Free for one project. One paid plan when you need more.

Free

Free forever

$0

Scope one real project and feed it to your agent.

  • 1 project
  • 100 features per project
  • 3 members per project
  • 1 MCP token per project
  • Unlimited open iterations
Create free project

Premium

Coming soon

$50per year

Two months free compared to monthly. Cancel any time.

  • 50 projects
  • 2000 features per project
  • 15 members per project
  • 10 MCP tokens per project
  • 20 open iterations
Coming soon

Stop briefing agents from memory.

One feature, one iteration, one line of config. Ten minutes.

Create free project