WebMCPify Core · Open source

Make your website usable by agents.

The local-first agent-readiness engine for new and existing applications. Discover what your product really does, expose controlled capabilities, test them in a browser, and independently verify the outcome.

webmcpify / local
$ webmcpify discover --path ./app
✓ routes and capabilities mapped

$ webmcpify review --path ./app
○ waiting for human approval

$ webmcpify test --path ./app \\
    --url http://localhost:3000
✓ outcome independently verified

Beyond code generation

Ground every capability in the real application.

Core starts from the implementation you have: routes, forms, APIs, handlers, state and authorization. It keeps generation subordinate to review, testing and proof.

01

Repository discovery

Understands the stack, routes, forms, handlers, APIs, state stores, authentication signals and existing WebMCP registrations.

02

Grounded capability design

Drafts strict tool schemas from behavior the application actually supports; generation is one controlled step inside the workflow.

03

Human-controlled change

Uses a disposable agent workspace and an approval manifest so proposed code cannot silently reach the target project.

04

Security checkpoint

Audits user and agent binding, backend authorization, origin scope, quotas, replay protection and input bounds before approval.

05

Real-browser testing

Runs approved tasks against the live site through Chrome and its experimental WebMCP inspection path.

06

Independent evaluation

Checks actual application state instead of trusting a tool or agent success message.

The Core workflow

Discover →Draft →Security check →Review →Apply →Test →Verify

07 CHECKPOINTS

Each stage has a distinct artifact or checkpoint. Approval and independent evidence stay visible all the way through.

  1. 01

    Discover

    Map routes, forms, APIs, handlers, state, auth signals and existing tools.

  2. 02

    Draft

    Generate grounded tools and browser-verifiable tasks in a disposable workspace.

  3. 03

    Security check

    Audit bindings, authorization, origin scope, quotas, replay protection and input bounds.

  4. 04

    Review

    Inspect the exact tools, tasks, security findings and source patch before approval.

  5. 05

    Apply

    Apply only an explicitly approved patch to the target checkout.

  6. 06

    Test

    Drive the live application through a real, WebMCP-capable browser agent.

  7. 07

    Verify

    Evaluate the application’s resulting state independently of the agent’s claim.

Install and run

Local by default.
Open source by design.

@improvisus/webmcpify-core is the open-source CLI and MCP workflow for creating, reviewing, testing, repairing and evaluating WebMCP capabilities.

Runtime Node.js 20.19+ on Node 20, 22.12+, or Node 23+Browser Chrome 150+ or a compatible Chromium build with WebMCP supportRepository Git with an initial commit and installed dependencies for build checksAgent An authenticated Codex, Claude Code, Gemini CLI, OpenCode, or Antigravity CLIOptional Temporal packages, service, and worker for durable final evaluation
npm install --global @improvisus/webmcpify-core
webmcpify run --url http://localhost:3000 --provider codex
webmcpify final-eval --path /path/to/target-project --url http://localhost:3000 --provider codex

Local MCP server

Bring Core into any coding agent.

Start the MCP server from the target repository. Its workspace boundary stays rooted there, and source changes still stop at the same review and approval workflow.

npx --yes --package @improvisus/webmcpify-core webmcpify-mcp
Read the complete setup ↗

SUPPORTED CLI AGENTS

Gemini CLI / Antigravity CLIClaude CodeCodexOpenCodeAny compatible CLI agent
Core detects an installed provider when --provider is omitted. Availability and end-to-end testing depend on the provider and local environment.