AI Web Action Standard (AWAS) is an open-source specification that enables AI browsers like Atlas, Comet, and other AI agents to interact with websites through machine-readable action definitions—without disrupting traditional human browsing.
Current AI browsers resort to slow, fragile browser automation that mimics human clicking. This approach:
- Is computationally expensive
- Breaks with UI changes
- Provides no semantic understanding
- Creates high server load
- Offers poor user experience
AWAS provides a dual-interface web architecture where:
- Human users browse normally through visual UI
- AI agents interact through structured, machine-readable actions
This is achieved through:
- AI Action Manifest - JSON specification of available actions
- HTML Data Attributes - Inline semantic hints for AI agents
- Extended Robots.txt - AI-specific policies and discovery
- Discovery Endpoints - Capability negotiation and metadata
- Server Middleware - Request handling and rate limiting
- ✅ Zero Breaking Changes - Works alongside existing HTML/CSS/JS
- ✅ Progressive Enhancement - Add incrementally, page by page
- ✅ Backwards Compatible - Traditional browsers unaffected
- ✅ Security First - Built-in rate limiting, auth, and permissions
- ✅ Open Standard - Royalty-free, community-driven
- ✅ Framework Agnostic - Works with any backend/frontend stack
- ✅ OpenAPI Interoperability - Link to existing OpenAPI/Swagger specs
AWAS is designed to be MCP (Model Context Protocol), A2A (Agent-to-Agent), ADK (Agent Development Kit), and OpenAPI/Swagger ready, enabling seamless integration with modern AI agent ecosystems and existing API infrastructure.
AWAS supports MCP through .well-known/mcp-manifest.json for standardized AI agent communication:
{
"name": "awas-site",
"version": "1.0.0",
"description": "AWAS-enabled site with MCP support",
"actions": {
"discover": "/.well-known/ai-actions.json",
"execute": "/api/ai-actions/{action_id}"
},
"capabilities": ["action-discovery", "action-execution", "state-management"]
}For AI Agents:
// Discover MCP capabilities
const manifest = await fetch('https://example.com/.well-known/mcp-manifest.json');
const actions = await fetch(manifest.actions.discover);AWAS can be used as a provider in agent-to-agent communication:
# ADK Integration Example
from agent_dev_kit import Provider
class AWASProvider(Provider):
def discover_actions(self):
return fetch_awas_manifest(self.base_url)
def execute_action(self, action_id, params):
return call_awas_action(action_id, params)AWAS actions can reference existing OpenAPI specifications to reduce duplication and leverage familiar tooling:
{
"id": "search_products",
"name": "Search Products",
"method": "GET",
"endpoint": "/api/search",
"openapi": {
"documentUrl": "/openapi.json",
"operationId": "searchProducts"
},
"inputSchema": {
"type": "object",
"properties": {
"q": {"type": "string"}
}
}
}Benefits:
- Link to existing OpenAPI/Swagger specs
- Reduce schema duplication
- Use familiar API documentation tools
- Maintain AWAS as the authoritative source for AI behavior
- Create AI Action Manifest (
/.well-known/ai-actions.json):
{
"specVersion": "1.1",
"lastUpdated": "2025-12-01T00:00:00Z",
"actions": [
{
"id": "search_products",
"name": "Search Products",
"description": "Search for products in the catalog",
"method": "GET",
"endpoint": "/api/search",
"intent": "read",
"sideEffect": "safe",
"conformanceLevel": "L1",
"openapi": {
"documentUrl": "/openapi.json",
"operationId": "searchProducts"
},
"parameters": [
{
"name": "query",
"type": "string",
"required": true,
"description": "Search query"
},
{
"name": "category",
"type": "string",
"enum": ["electronics", "clothing", "books"],
"description": "Filter by category"
}
],
"inputSchema": {
"type": "object",
"properties": {
"query": { "type": "string" },
"category": { "type": "string", "enum": ["electronics", "clothing", "books"] }
},
"required": ["query"]
},
"outputSchema": {
"type": "object",
"properties": {
"results": {
"type": "array",
"items": {
"type": "object",
"properties": {
"id": { "type": "string" },
"name": { "type": "string" },
"price": { "type": "number" }
}
}
}
}
}
}
]
}
- Add HTML Data Attributes (optional enhancement):
<div data-awas-action="search_products" data-awas-element="search-form">
<input type="text" name="query" data-awas-param="query">
<select name="category" data-awas-param="category">
<option value="electronics">Electronics</option>
<option value="clothing">Clothing</option>
</select>
<button data-awas-trigger>Search</button>
</div>- Update robots.txt:
User-agent: *
Allow: /
# AI Agent Directives
AI-Action-Manifest: /.well-known/ai-actions.json
AI-Rate-Limit: 60/minute
AI-Auth-Required: Bearer
const AWAS = require('@awas/client');
const client = new AWAS.Client('https://example.com');
// Discover available actions
const manifest = await client.discover();
// Execute an action
const results = await client.executeAction('search_products', {
query: 'laptop',
category: 'electronics'
});
console.log(results);- Better AI Rankings - Structured data helps AI understand your site
- Reduced Server Load - Efficient API calls vs. page scraping
- Control & Visibility - Define what AI can and cannot do
- Future-Proof - Adapt to AI-driven web without redesign
- Competitive Advantage - Early adopters rank higher in AI results
- Faster Execution - Direct action calls vs. DOM automation
- Reliable Operation - Resilient to UI changes
- Rich Semantics - Understand intent, not just pixels
- Standard Protocol - One integration works everywhere
- Faster Results - AI completes tasks in seconds
- More Reliable - Fewer failures and retries
- Better Privacy - Explicit permissions and audit trails
- Improved Experience - Seamless AI assistance
AWAS includes security features:
- Rate limiting specifications
- Authentication requirements
- Action permission system
- Audit trail capabilities
- Privacy-respecting directives
- CSRF protection guidelines
See SECURITY.md for details.
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
- Report bugs via GitHub Issues
- Suggest features via Discussions
- Submit pull requests for improvements
AWAS is licensed under the Apache License, Version 2.0.
You must include NOTICE and this license when redistributing or creating derivative works.
See LICENSE and NOTICE for details.
This project was formerly MIT licensed; as of October 29, 2025, AWAS is under Apache 2.0.
- Inspired by Schema.org, OpenAPI, and WAI-ARIA standards
- Built for compatibility with MCP (Model Context Protocol) and A2A (Agent2Agent)
- Community feedback from AI browser developers and web developers
- Issues: GitHub Issues
- Discussions: GitHub Discussions
- Website: Coming soon
- v1.0 - Core specification and basic examples
- v1.1 - Enhanced discovery, typing, and safety contracts (specVersion, conformance levels L1/L2/L3, typed I/O)- [ ] v1.2 - Real-time update capabilities
- v1.1 - MCP, A2A/ADK, and OpenAPI/Swagger interoperability
- v1.3 - Advanced authentication mechanisms
- v2.0 - Extended protocol integrations
Made with ❤️ for the AI-driven web