LAYGENT_

Your app already knows
its actions.
Laygent exposes them
to AI agents.

Laygent automatically generates MCP tooling and semantic interfaces from your app's components, actions, and state. So agents can interact natively — not through browsers.

$ npx laygentify> View Generated Tools
laygent@cli:~
$ npm install laygent
$ laygentify

✓ scanning components...
✓ extracting actions...
✓ generating MCP tools...
✓ generating semantic runtime...
✓ generating permissions...
✓ agent interface ready.

$
React App
↓
Components + Actions + State
↓
Laygent
↓
MCP
Permissions
Tool Graph
↓
AI Agents
→Your existing application
→Laygent reads your app's runtime structure
→Laygent builds a secondary machine-native interaction layer
→Structured tools, permissions, and relationships for agents
→Agents interact natively through tools, not UI
</>
Define Once

Wrap actions, state, and components in your app as usual.

⚡
Auto Generate

Laygent generates MCP tools, schemas, and tool graphs.

⛨
Secure by Default

Permissions, access control, and audit logs built-in.

>_
Agent Ready

Agents interact through semantic tools, not brittle browsers.

> Laygent creates a secondary machine-native interaction layer on top of existing applications.
// 01 the problem

Stop forcing agents to use software through pixels.

Two stacks. One is duct tape pretending to be infrastructure. The other is a runtime.

// today — agent automation
  1. [1]AI Agent×
    ↓
  2. [2]Playwright×
    ↓
  3. [3]DOM Parsing×
    ↓
  4. [4]OCR / Screenshots×
    ↓
  5. [5]Button Clicking×
    ↓
  6. [6]Breaks Constantly×
↯ collapses on every UI change
// laygent — agent-native runtime
  1. [1]AI Agent●
    ↓
  2. [2]Semantic Actions●
    ↓
  3. [3]Generated MCP Tools●
    ↓
  4. [4]Direct App Runtime Access●
    ↓
  5. [5]Reliable Execution●
✓ deterministic, typed, permissioned
> Agents do not want pixels.
> They want actions, state, permissions, and workflows.
// 02 developer api

One declaration. One tool. Zero selectors.

Wrap an action once. Laygent produces a typed, permissioned MCP tool any agent can call.

actions/refund.ts
agent.action("approve_refund", {
  input: RefundSchema,
  permissions: ["admin"],
  handler: approveRefund,
})
// generated tool
approve_refund(refund_id)
// instead of
page.click("#approve")
// 03 philosophy

Browser automation is temporary infrastructure.

AI agents are not humans. They should not interact with software like humans do.

// agents should NOT depend on
  • ×pixels
  • ×screenshots
  • ×DOM trees
  • ×brittle selectors
  • ×fake mouse clicks
// agents should interact through
  • ✓semantic actions
  • ✓state
  • ✓workflows
  • ✓permissions
  • ✓structured runtime interfaces
// 04 pitch
Problem
Agents break because software was built only for humans.
Solution
Laygent generates agent-native interfaces from existing apps.
Market
Every SaaS product will eventually need an AI interaction layer.
Wedge
Start with React/Next.js internal tools and SaaS dashboards.
Vision
React components become executable agent tools.
// ready when you are
$ laygentify your app
npm install laygent · MIT · zero runtime config