WORK IN PROGRESS — We're actively building. Things may change.

Drop into your app, write some WebMCP tools to give it hands, and become AI-native.

Two script tags. No backend rewrite. Your users will think you built it that way.

See it in action

Ask the agent to register a new tool or change its appearance. Watch the panels update in real-time.

Char Demo
Live

No tools registered

Ask the agent to create one

Agent

Loading agent...

Try: "Register a joke tool" or "Change to purple theme"

The Problem
Complexity

We're overcomplicating agents.

Most frameworks want you to build from scratch—new APIs, new backends, custom UIs just to approve tool calls. All that code to mimic a fraction of what users already do in your UI.

New API Layer
Custom Backend
Approval UI
Months of work
Safety

Too dangerous to ship.

Service accounts with broad access. Credentials scattered across systems. Audit logs that say "Agent" instead of "Sarah from Sales." That's not enterprise security—that's a breach waiting to happen.

Svc Acct admin@sys API_KEY=sk-... DB_PASS=*** Audit Log 12:04 "Agent" → write 12:05 "Agent" → delete 12:06 "???" → export ? ?

Char solves both.

Pillar 1

Start where the logic lives.

Most agent frameworks start with backend APIs, then struggle to connect them to your UI. Char flips this. WebMCP tools are thin wrappers around actions your frontend already handles. Start there. Need server-side tools later? Add any MCP server. Open protocols all the way down.

index.html
WebMCP MCP
<!-- That's it. -->

<script src="char.js"></script>
<char-agent />
Pillar 2

Knowledge that compounds.

Skills are how your org teaches Char. Any user can create them, refine them, or ask the agent to capture a workflow. They're shared across your organization—every improvement makes the agent better for everyone. Expertise stops walking out the door.

process-return.skill.md ORG-WIDE
# Created by @sarah, refined by @mike
name: process-return
description: Handle return requests
## Steps
1. Verify order exists and belongs to user
2. Check eligibility (30 days, unused)
3. Generate return label via shipping API
Pillar 3

Where power meets
accountability.

Every Tool Hub is scoped to a single identity. Every action flows through your IDP. Every invocation is logged, governed, and revocable.

Not because we wanted enterprise customers. Because it's the only way to build this responsibly.

Your IDP
The only authority
Instant Revocation
One place to end access
Your IDP
Tool Hub
S
sarah@acme.com
Salesforce
sarah
Database
sarah
Internal API
sarah
Audit
sarah → read
sarah → write
sarah → query
S

Disable Sarah in Okta → access ends everywhere

Built to Ship

The agent writes it.
The agent tests it.

Traditional tools fail because developers assume one thing, AI assumes another. Char's Chrome DevTools MCP lets Claude test WebMCP tools the moment they're written—same AI that will use them in production.

Schema mismatches, missing error handling, unclear descriptions—all caught before you ship.

Same AI
Writes and tests the tools
Real Browser
Tests in actual Chrome
Instant Feedback
Catches issues immediately
Ship Confident
If it works, it ships
Same AI Claude Code (Write) Chrome DevTools MCP (Test) If it works It ships
Ready to Start

Add AI to your app today.

Set up in minutes with Claude Code. The AI writes the tools, tests them, and deploys. No integration project required.

+ TDD for AI - tested before production
+ No backend rewrite needed
+ Works with any framework
+ Standards-first (MCP, WebMCP)

Join the Waitlist

Get early access and shape the future of AI-powered apps.

MCP compatible • WebMCP v1 • BYO inference