WebMCP Readiness Audit
Grade your MCP server for Claude in Chrome readiness using a 6-tier audit.
As Claude moves into web browsers, MCP servers must meet higher standards. This audit tests your server against the WebMCP standard for AI-native web automation.
6-Tier Evaluation Framework
Each tier tests a different part of MCP server readiness. Scores give more weight to core features than advanced ones.
Server Implementation
Core MCP protocol rules and response handling
- Server responds to MCP requests
- Latest protocol version (2024-11-05)
- JSON-RPC error handling
- Response time < 5s
Tool Discoverability
Tools exposed with proper schemas and descriptions
- Tools exposed via tools/list
- Descriptive tool descriptions
- Input schemas defined
- Property descriptions
Instrumentation
Monitoring, health checks, and OAuth metadata
- /health endpoint
- /info endpoint with metadata
- OAuth Protected Resource metadata
Consistency
Session handling, saved state, and recovery
- Session management tools
- Self-healing/recovery tools
- Consistent response structure
Agent Optimizations
Vision, natural language, and cognitive persona support
- Screenshot/vision capabilities
- Natural language inputs
- Persona-based testing
- Accessibility testing
Documentation
AI-readable docs and machine-friendly metadata
- /llms.txt or /webmcp.txt
- Documentation endpoint
Grade Scale
Your overall score is calculated from weighted tier scores. Here's what each grade means:
Common Issues & Remediation
The most common issues found during WebMCP audits, sorted by impact.
| Severity | Issue | Tier | Remediation |
|---|---|---|---|
| CRITICAL | Server not responding to MCP requests | Tier 1 | Verify MCP endpoint is accessible and returns 200 OK |
| MEDIUM | Old protocol version | Tier 1 | Upgrade to protocol version 2024-11-05 |
| CRITICAL | No tools exposed | Tier 2 | Implement and register at least one MCP tool |
| HIGH | Tools lack input schemas | Tier 2 | Add JSON Schema for all tool inputs |
| MEDIUM | Missing descriptions | Tier 2 | Add help text to all tools (at least 10 characters) |
| MEDIUM | No /health endpoint | Tier 3 | Add GET /health endpoint for monitoring |
| LOW | No OAuth metadata | Tier 3 | Add /.well-known/oauth-protected-resource for Claude.ai |
| MEDIUM | No session management | Tier 4 | Add session save/load tools for keeping state |
| LOW | No self-healing | Tier 4 | Add selector healing or retry tools |
| MEDIUM | No vision/screenshot support | Tier 5 | Add screenshot tool for visual verification |
| MEDIUM | No /llms.txt file | Tier 6 | Create /llms.txt with AI-readable server documentation |
Quick Start
Audit any MCP server with a single command.
npx cbrowser webmcp-ready https://your-server.com/mcp
npx cbrowser webmcp-ready https://your-server.com/mcp --api-key YOUR_KEY
npx cbrowser webmcp-ready https://your-server.com/mcp --html
MCP Tool
Use the webmcp_ready_audit MCP tool directly from Claude Desktop or Claude.ai.
webmcp_ready_audit
Audit an MCP server for Claude in Chrome / WebMCP readiness. Uses 6-tier scoring: Server (25%), Tools (20%), Instrumentation (15%), Consistency (15%), Agent Features (15%), Docs (10%).
Pro Tips
Start with Tier 1
Fix critical server issues first. Without a working MCP endpoint, agents cannot use any of your tools.
Add llms.txt
Adding an llms.txt file is a quick win. It helps agents learn your server's features before they call any tools.
OAuth for Claude.ai
Want your server as a Claude.ai custom connector? Add OAuth Protected Resource metadata at /.well-known/oauth-protected-resource.
Test with CBrowser
Use CBrowser's demo server as a reference. Run the audit on https://demo.cbrowser.ai/mcp to see what an A-grade server looks like.
Ready to Test Your MCP Server?
Run a WebMCP audit to see your score. Then follow the tier-by-tier steps to improve readiness.