Skip to main content
The most powerful feature of Char isn’t the embedded agent itself—it’s the development workflow that guarantees your tools work before users ever see them.

​
The Problem with AI Tools

Most AI integrations follow this pattern:
  1. Developer writes tool code
  2. Developer manually tests the tool
  3. AI agent tries to use the tool in production
  4. Users discover bugs the developer missed
The AI and the developer have different mental models. What seems obvious to a human (“click the submit button”) may not match what the AI actually needs (“dispatch a click event on the element with data-testid=‘submit’”).

​
The Solution: Same Tools, Same AI

Char uses WebMCP—a standard browser API for registering tools that AI agents can call. What makes this powerful is that the same AI that writes your tools also tests them. When you use Claude Code to set up Char:
  1. Claude writes a tool — e.g., fill_login_form
  2. Claude calls that tool — via Chrome DevTools MCP to verify it works
  3. The embedded agent uses the same tool — no translation, no mismatch

​
Why This Matters

​
No Hope-Based Development

Traditional integrations hope the AI can figure out vague descriptions. With Char, if Claude can’t successfully call a tool during development, it knows immediately and fixes the implementation.

​
Same Interface, Zero Translation

The WebMCP API that Claude tests is identical to what the embedded agent uses. There’s no “works on my machine” gap between development and production.

​
AI Dogfooding

Your tools are tested by AI before users see them. If Claude struggles to use a tool, your users’ AI would struggle too. Claude catches these issues during setup.

​
The Standard

WebMCP is a W3C proposal for navigator.modelContext—a browser API that lets websites expose tools to AI agents. Char’s embedded agent and Chrome DevTools MCP both consume this same standard. This means:
  • Tools you write for Char work with any WebMCP-compatible agent
  • Testing with Claude Code validates the same interface your users experience
  • The ecosystem grows together as more tools adopt the standard

​
Further Reading

WebMCP Documentation

Full API reference and examples

Chrome DevTools MCP

The testing bridge for Claude Code