@ashpfeif12/agentweb
v0.1.0
Published
Make the web agent-ready. Open spec, scorer, and MCP generator for the agent-driven web.
Maintainers
Readme
🌐 AgentWeb
Make the web agent-ready.
AgentWeb is an open platform for bridging the browser-centric web and the emerging agent-driven web. It provides a spec, tools, and infrastructure so AI agents can discover, query, and transact with any website or service.
# Score any site's agent readiness in 30 seconds
npx agentweb score https://yoursite.com
# Generate an MCP server from your existing API
npx agentweb generate --from openapi.yamlThe Problem
The web was built for browsers. AI agents are the next consumer — but they can't discover what your site does, what actions are available, or how to interact with it. There's no robots.txt for agents. No structured way to say "here's what I offer and how to use it."
Result: Every agent integration is bespoke. Every brand is invisible to agents. The infrastructure gap is massive.
The Solution: Five Layers
| Layer | What | Type | |-------|------|------| | agent.json | Open spec describing what your site offers to agents | 🟢 Open Source | | MCP Generator | Auto-generate MCP servers from existing APIs | 🟢 Open Source | | Agent Middleware | Proxy layer making any site agent-consumable | 🔵 Commercial | | Agent Commerce | Structured catalogs & policies for agent negotiation | 🔵 Commercial | | Agent SEO | Analytics for how agents discover your brand | 🔵 Commercial |
Open source bottom creates adoption. Commercial top solves enterprise problems.
📋 agent.json — The Spec
Drop an agent.json at your domain root. It tells agents what you do and how to interact:
{
"name": "Acme Fashion",
"version": "1.0.0",
"spec_version": "1.0",
"description": "Premium fashion retailer — 2,000+ products across women's and men's collections",
"capabilities": [
{
"name": "search_products",
"description": "Find products by category, color, size, price range, or occasion",
"type": "query"
},
{
"name": "place_order",
"description": "Add items to cart and complete purchase",
"type": "action"
}
],
"authentication": {
"type": "oauth2",
"authorization_url": "https://acmefashion.com/oauth/authorize",
"scopes": ["read_catalog", "place_orders"]
},
"policies": {
"rate_limit": "100/minute",
"returns": "30-day no-questions-asked",
"data_handling": "no_training_on_interactions"
},
"brand_voice": {
"tone": "warm, knowledgeable, never pushy",
"prohibited": ["competitor comparisons", "unauthorized discounts"]
},
"endpoints": {
"mcp": "https://mcp.acmefashion.com/sse",
"rest": "https://api.acmefashion.com/v2",
"docs": "https://developers.acmefashion.com"
}
}📊 Readiness Scorer
Score any website on agent readiness (0–100):
npx agentweb score https://example.com┌─────────────────────────────────────────┐
│ Agent Readiness Report: example.com │
├─────────────────────────────────────────┤
│ Discovery ██████░░░░░░ 14/25 │
│ Structure ████████░░░░ 18/25 │
│ Actions ██░░░░░░░░░░ 6/25 │
│ Policies █░░░░░░░░░░░ 4/25 │
├─────────────────────────────────────────┤
│ TOTAL SCORE 42/100 │
│ │
│ Top recommendations: │
│ 1. Add agent.json manifest │
│ 2. Define action schemas │
│ 3. Add machine-readable policies │
└─────────────────────────────────────────┘⚡ MCP Server Generator
Generate a fully functional MCP server from your existing API:
# From OpenAPI spec
npx agentweb generate --from openapi.yaml --output ./my-mcp-server
# From GraphQL schema
npx agentweb generate --from schema.graphql --output ./my-mcp-server
# From a live website (experimental)
npx agentweb generate --from https://example.com --output ./my-mcp-serverOutputs a complete, deployable package: MCP server code, Dockerfile, agent.json, tests, and README.
🏗 Project Structure
agentweb/
├── packages/
│ ├── spec/ # agent.json JSON Schema + TypeScript types + validation
│ ├── scorer/ # CLI readiness scorer
│ └── generator/ # API → MCP server generator
├── examples/ # Example agent.json files by industry
├── apps/
│ └── docs/ # Documentation site
└── .github/
└── workflows/ # CI/CDContributing
We welcome contributions! The spec is designed to evolve with community input.
- 🐛 Bug reports: Open an issue
- 💡 Spec proposals: Start a discussion
- 🔧 Code contributions: See CONTRIBUTING.md
License
packages/specandpackages/scorerandpackages/generator: MIT- Commercial products: Proprietary
