create

The create command is the core of MCPKit. It analyzes a website and generates a complete MCP server with tools for interacting with that site.

Usage

mcpkit create <url> [options]

Arguments

string

required

The URL of the website to create an MCP server for. Must be a valid HTTP or HTTPS URL.

Options

boolean

default:"false"

Skip the authentication step even if the website requires login.

Examples

Basic Usage

Create an MCP server for a public website:

mcpkit create https://news.ycombinator.com

With Interactive URL Prompt

If you don’t provide a URL, mcpkit will prompt you:

mcpkit create
# ? Enter the URL of the website to create an MCP for: https://mcpkit.sh

Skip Authentication

For testing or public websites:

mcpkit create https://example.com --skip-auth

How It Works

URL Validation

MCPKit validates the provided URL and extracts the domain name.

🔨 MCP Server Generator
📍 Analyzing: https://mcpkit.sh

Browser Session

A headless browser session is launched via Browserbase to load the website.

The browser runs in the cloud, so you don’t need Chrome installed locally.

Authentication (Optional)

If the website requires authentication, you’ll be prompted to log in:

🔐 This site may require authentication.
Would you like to authenticate? (Y/n) y

🌐 Opening live browser session...
Complete authentication in your browser, then press Enter...

Your authentication context (cookies, session data) will be saved for future use.

Action Discovery

MCPKit uses AI to analyze the page and discover available actions:

🔍 Discovering actions...
✅ Found 8 actions:
  - Create new issue
  - Search issues
  - Update issue status
  - Add comment
  - List projects
  - ...

The AI examines:

Interactive elements (buttons, forms, links)
Page structure and navigation
Common workflows and patterns
API endpoints (if available)

Schema Generation

For each discovered action, MCPKit generates:

Tool name and description
Input parameters with types
Zod validation schemas
Implementation code

// Generated tool example
{
  name: "create_issue",
  description: "Create a new issue",
  inputSchema: {
    type: "object",
    properties: {
      title: { type: "string" },
      description: { type: "string" },
      priority: {
        type: "string",
        enum: ["low", "medium", "high", "urgent"]
      }
    },
    required: ["title"]
  }
}

Server Generation

A complete MCP server project is created:

📁 mcp-stagehand-mcpkit.sh/
├── src/
│   ├── index.ts           # Main MCP server
│   ├── tools/
│   │   ├── create_issue.ts
│   │   ├── search_issues.ts
│   │   └── ...
│   └── types.ts           # Shared types
├── package.json
├── tsconfig.json
└── README.md

Your MCP server is ready to use!

Generated Server Structure

The generated MCP server includes:

Main Server (`src/index.ts`)

import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";

const server = new Server(
  {
    name: "hackernews-mcp-server",
    version: "1.0.0",
  },
  {
    capabilities: {
      tools: {},
    },
  }
);

// Tool handlers
server.setRequestHandler(ListToolsRequestSchema, async () => ({
  tools: [
    // Generated tools
  ],
}));

server.setRequestHandler(CallToolRequestSchema, async (request) => {
  // Tool implementation
});

Tool Implementation

Each tool is implemented with:

// tools/create_issue.ts
export async function createIssue(params: {
  title: string;
  description?: string;
  priority?: "low" | "medium" | "high" | "urgent";
}) {
  const stagehand = await initStagehand();

  // Navigate to create issue page
  await stagehand.act("click new issue button");

  // Fill in form
  await stagehand.act(`type "${params.title}" in title field`);

  if (params.description) {
    await stagehand.act(`type "${params.description}" in description field`);
  }

  // Submit
  await stagehand.act("click create button");

  // Extract result
  const result = await stagehand.extract("get created issue details");

  return result;
}

Configuration

Generated servers can be configured via environment variables:

# .env in generated server directory
BROWSERBASE_API_KEY=your_api_key
BROWSERBASE_PROJECT_ID=your_project_id
LOG_LEVEL=info

Testing the Generated Server

After creation, test your server:

cd mcp-stagehand-<domain>/
npm install
npm run build
npx @modelcontextprotocol/inspector node dist/index.js

Advanced Options

Custom Model for Discovery

The AI model used for action discovery can be configured in your ~/.mcpkit/secrets.json:

{
  "llmProvider": "google",
  "llmApiKey": "your-api-key",
  "llmModel": "gemini-2.0-flash-exp"
}

Supported providers:

google - Gemini models (recommended)
openai - GPT models
anthropic - Claude models
azure - Azure OpenAI

Debugging

For verbose output during generation:

DEBUG=mcpkit:* mcpkit create https://example.com

Common Workflows

Public Website
Authenticated Website
Internal Tool

For websites that don’t require authentication:

# Create server
mcpkit create https://news.ycombinator.com

# Test it
cd mcp-stagehand-news.ycombinator.com
npm install && npm run build
npx @modelcontextprotocol/inspector node dist/index.js

For websites requiring login:

# Create server with auth
mcpkit create https://mcpkit.sh
# Follow prompts to authenticate

# Context is saved, regenerate without auth:
mcpkit create https://mcpkit.sh --skip-auth

For company internal tools:

# Make sure you can access the tool
mcpkit create https://internal.company.com

# You may need VPN or network access

Ensure you have permission to automate interactions with internal tools.

Troubleshooting

No actions discovered

If MCPKit doesn’t find any actions:

Try a more specific URL - Navigate to a specific page with clear actions
# Instead of homepage mcpkit create https://mcpkit.sh/issues

Authenticate first - Some content only appears after login

mcpkit create https://example.com
# Choose Yes when prompted to authenticate

Check the website is accessible - Make sure it’s not behind a firewall or paywall

Authentication fails

If authentication doesn’t work:

Complete the full login flow - Don’t close the browser until you see success
Check for 2FA - Some sites require two-factor authentication
Verify credentials - Make sure you’re using valid credentials
Try manual context creation:
mcpkit contexts create example.com

Generated server doesn't work

If the generated tools don’t work correctly:

Check Browserbase API key in the generated .env file
Verify authentication - Make sure saved context is still valid
Test individual actions in the MCP Inspector

Regenerate with updated context:

mcpkit contexts delete example.com
mcpkit create https://example.com

Next Steps

Manage Contexts

Learn how to manage authentication contexts

Configure Secrets

Update your API keys and settings

View Examples

See example generated servers

Getting started

CLI Commands

Examples

Advanced

Usage

Arguments

Options

Examples

Basic Usage

With Interactive URL Prompt

Skip Authentication

How It Works

Generated Server Structure

Main Server (`src/index.ts`)

Tool Implementation

Configuration

Testing the Generated Server

Advanced Options

Custom Model for Discovery

Debugging

Common Workflows

Troubleshooting

Next Steps

Manage Contexts

Configure Secrets

View Examples

​Usage

​Arguments

​Options

​Examples

​Basic Usage

​With Interactive URL Prompt

​Skip Authentication

​How It Works

​Generated Server Structure

​Main Server (src/index.ts)

​Tool Implementation

​Configuration

​Testing the Generated Server

​Advanced Options

​Custom Model for Discovery

​Debugging

​Common Workflows

​Troubleshooting

​Next Steps

Manage Contexts

Configure Secrets

View Examples

Usage

Arguments

Options

Examples

Basic Usage

With Interactive URL Prompt

Skip Authentication

How It Works

Generated Server Structure

Main Server (`src/index.ts`)

Tool Implementation

Configuration

Testing the Generated Server

Advanced Options

Custom Model for Discovery

Debugging

Common Workflows

Troubleshooting

Next Steps