WebMCPify

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for WebMCPify

WebMCPify makes web applications ready for AI agents while keeping humans in control of every source change. It is a domain-agnostic CLI and local MCP workflow for turning new applications, existing applications, and applications with partial WebMCP integrations into reliable, testable WebMCP experiences. Instead of asking an agent to guess how a website works from its interface, WebMCPify analyzes the application and creates grounded tools from its real routes, UI actions, APIs, state, authentication flows, and existing WebMCP signals. The workflow is: discover → baseline → generate → review → approve → apply → WebMCP test → repair → validate Discovery maps the target application and records its relevant files, routes, capabilities, forms, state, APIs, and existing integrations. A baseline browser evaluation measures what the application can already do before any changes are made. WebMCPify then uses its existing generation pipeline and configured AI provider to propose WebMCP registrations. Generated changes are created in an isolated workspace and returned as an inspectable patch. The human reviews the proposed tools, tasks, files, and diff before approving anything. Only the approved patch can be applied to the target repository. After application, WebMCPify runs browser-based tests using headless Google Chrome, Chrome DevTools Protocol, and playwright-core. The tests verify that agents can discover and use the tools against the live application. Verification is based on observable browser state and task assertions rather than an agent simply claiming success. If a test fails, WebMCPify can create a repair proposal. Repair is also subject to human approval, so an automated repair cannot silently modify the project. Git fingerprints, patch boundaries, approval manifests, build checks, isolated workspaces, and rollback protections help prevent unsafe or unintended changes. For durable repair and final evaluation, WebMCPify integrates Temporal. Temporal preserves workflow state, coordinates long-running evaluation steps, and allows repair and validation runs to continue reliably. The final evaluation compares the baseline, WebMCP-enabled result, and Temporal-backed result using independent task verification and saved trajectory evidence. WebMCPify is available in multiple ways:

• As a CLI for humans and coding agents.
• As a local MCP server that exposes repository analysis, generation, approved application, and testing tools.
• As a published npm package for global installation or npx usage.
• As a WebMCP-native demo page that exposes tools directly to compatible browser-based agents. The MCP server operates against the agent’s current local repository and does not require GitHub OAuth or repository uploads. It is started from the target project so the workspace boundary is clear. Coding agents can use tools such as analyze_repository, generate_webmcp, apply_webmcp, and test_webmcp, while humans retain control over permissions, source changes, patch approval, repairs, and deployment. The project follows the current WebMCP direction, including document.modelContext.registerTool(), tool metadata, structured input schemas, lifecycle signals, and browser-native tool registration. It is designed to support practical WebMCP development today while remaining useful as the platform evolves. WebMCPify helps teams prototype agent-ready applications faster, bring existing sites into the WebMCP ecosystem, and continuously test and improve integrations with evidence instead of guesswork. Demo: https://webmcpify-homepage.vercel.app/ Repository: https://github.com/abeebridwan/webmcpify

Built With

• agents
• ai
• automation
• browser
• chrome
• coding
• context
• devtools
• git
• google
• model
• next.js
• node.js
• npm
• playwright
• protocol
• react
• temporal
• typescript
• vercel
• webmcp