Quick Start

Get your first agent running in under 5 minutes. No installation required.

Prerequisites

• Modern browser — Chrome 90+, Firefox 90+, Safari 15+, Edge 90+
• API key (optional) — For live mode, get one at console.anthropic.com

The builder works entirely in your browser. No server, no installation, no sign-up. Your API key stays in localStorage and goes directly to the Anthropic API.

Agent Builder

The Agent Builder is a split-panel interface for configuring and testing AI agents. The left panel holds the config form, the right panel is a live chat preview.

Creating an Agent

Every agent starts with a config object that defines its behavior. The builder provides a visual form for creating this config. You can also fork existing agents from the marketplace.

Minimum required fields: name, system_prompt, and model. Everything else has sensible defaults.

Config Reference

Personality

The system_prompt field defines your agent's core behavior. It's sent as the system parameter in every API call. Write it like you're briefing a new team member.

# Coding Assistant Agent

You are an expert software engineer. When given a task:
1. Analyze the requirements thoroughly
2. Consider edge cases and error handling
3. Write clean, well-documented code
4. Suggest tests for critical paths

Languages: TypeScript, Python, Rust, Go

# Research Agent

You are a research coordinator. Break complex questions
into sub-tasks and search multiple sources.

Process:
1. Decompose the question into 3-5 search queries
2. Search academic papers, news, and technical docs
3. Cross-reference findings for accuracy
4. Synthesize into a clear summary with citations

# DeFi Analyst Agent

You are a DeFi protocol analyst. Monitor on-chain data,
analyze yield opportunities, and assess risk.

Capabilities:
- Pool liquidity analysis (Uniswap, Raydium, Jupiter)
- APY calculation with impermanent loss modeling
- Smart contract risk assessment
- Gas cost optimization recommendations

Tools

Tools are conceptual capabilities you assign to your agent. In demo mode, tools influence which simulated response patterns match. In live mode, the selected tools are listed in the system prompt context so the model knows what capabilities to reference.

Saving & Sharing

Save to Browser

Click Save Agent to persist your config to localStorage under the key clawdpad_agents. Saved agents appear in the My Agents section of the marketplace panel, where you can load, share, or delete them.

Agents are saved by handle — saving with the same handle overwrites the previous version.

Export JSON

Click Export JSON to download the agent config as a .json file. This is a portable backup that includes all fields.

{
  "name": "ResearchBot",
  "handle": "@research_bot",
  "system_prompt": "You are a research assistant...",
  "model": "claude-sonnet-4-5-20250929",
  "tools": ["web_search", "code_exec"],
  "visibility": "public",
  "created": "2026-02-17T12:00:00.000Z"
}

Share via URL

In the My Agents section, click Share on any saved agent. This encodes the config as a base64 URL hash and copies it to your clipboard.

The URL format is: clawdpad.com/#import=<base64-encoded-config>

Import

When someone opens a ClawdPad URL with an #import= hash, the config is automatically decoded and loaded into the builder. The hash is then cleared from the URL bar.

You can also import by forking any agent from the marketplace.

Marketplace

The marketplace contains 24 preset agents across 7 categories, plus any agents you've saved locally.

Browsing

• Search — Full-text search across agent names, handles, descriptions, and tags
• Category pills — Filter by DeFi, Security, Research, Content, Dev Tools, Analytics, or Automation
• Sort — Trending (default), Newest, Most Forked, or Top Rated

The footer shows how many agents match your current filters.

Forking

Click Fork to Builder on any agent card to load its full config into the builder. The name gets a "(fork)" suffix. Modify any field before saving.

Try It

In the agent detail modal, click Try It to fork the agent and immediately start a conversation. In demo mode, the agent responds using personality-aware simulated messages. In live mode (with API key), you get real streaming responses.

Categories

API Integration

How It Works

ClawdPad makes direct browser-to-API calls to the Anthropic Messages API. There is no backend proxy or middleware.

1. You enter your Anthropic API key via the Key button in the chat bar
2. The key is stored in localStorage (never sent to ClawdPad servers)
3. When you send a message, ClawdPad calls POST https://api.anthropic.com/v1/messages with stream: true
4. Streaming SSE responses are parsed in real-time and rendered with markdown formatting
5. The anthropic-dangerous-direct-browser-access header is set to enable CORS

POST https://api.anthropic.com/v1/messages
Headers:
  x-api-key: sk-ant-...
  anthropic-version: 2023-06-01
  anthropic-dangerous-direct-browser-access: true
  Content-Type: application/json

Body:
  model: (from agent config)
  system: (agent's system_prompt)
  messages: (conversation history)
  max_tokens: 4096
  stream: true

Supported Models

Security

• Your API key is stored only in your browser's localStorage
• API calls go directly from your browser to api.anthropic.com
• No ClawdPad server ever sees your key or messages
• You can remove your key at any time via the Key button → Remove
• Keys are validated on save with a lightweight API call

MCP Integration (Roadmap)

Future versions of ClawdPad will support exposing agents as MCP servers for use with Claude Code, Cursor, and other MCP-compatible tools.

{
  "mcpServers": {
    "clawdpad-agent": {
      "command": "npx",
      "args": ["clawdpad-mcp", "--agent", "@my_agent"]
    }
  }
}

This is not yet implemented. The config format above is aspirational and subject to change.

FAQ