Web Agent Protocol
Emerging protocols defining how AI agents securely interact with websites.
Web Agent Protocols are a collection of emerging standards that define how AI agents can communicate with websites and web services in a structured and secure way. They form the infrastructure layer for the next generation of AI-web interaction.
Examples of protocols
MCP — bibliotheekterm (Model Context Protocol): Anthropic's standard for AI tool communication. A2A (Agent-to-Agent): Google's protocol for communication between AI agents. Web Bot Auth: authentication standard for bots. These protocols are developing rapidly and being adopted more widely.
Why this matters for websites
Websites that adopt these protocols early are more discoverable and usable for AI agents. Just as websites needed to become mobile-ready in the early 2000s, they now need to become agent-ready. The scanner measures how far a website is on this path.
Comparison: MCP vs A2A vs Web Bot Auth
| Feature | MCP (Model Context Protocol) | A2A (Agent-to-Agent) | Web Bot Auth |
|---|---|---|---|
| Developer | Anthropic | Google DeepMind | Community/W3C proposal |
| Primary purpose | Connect AI model with tools and data | Communication between autonomous AI agents | Standardized authentication for bots |
| Communication model | Client-server (AI client calls MCP server) | Peer-to-peer (agent-to-agent) | Client-server (bot authenticates with service) |
| Transport | HTTP (Streamable) / SSE / stdio | HTTP with JSON-RPC | HTTP with OAuth 2.0 |
| Discovery | /.well-known/mcp.json, meta tags, DNS | /.well-known/agent.json (Agent Card) | /.well-known/oauth-authorization-server |
| Authentication | OAuth 2.0 (optional) | OAuth 2.0, API keys | OAuth 2.0 (core) |
| Status (2026) | Widely adopted, open source | Growing adoption, open specification | Early stage, in development |
| Use case | AI tools, databases, APIs, web services | Multi-agent workflows, task delegation | Secured bot access to web services |
| Open source | Yes (Apache 2.0) | Yes | Yes (proposal) |
How these protocols work together
The three protocols complement each other in the AI-web ecosystem:
- Web Bot Auth handles authentication: how does an AI agent prove it's authorized?
- MCP handles interaction: how does the agent use a website's tools and data?
- A2A handles coordination: how do multiple agents collaborate on a complex task?
A typical scenario: an AI agent uses Web Bot Auth to log in, MCP to search products and place an order, and A2A to engage a payment agent for the transaction.
What does our scanner check?
The scanner checks your website for agent-readiness signals: MCP discovery (/.well-known/mcp.json), OAuth Discovery — bibliotheekterm endpoints, and other protocol-related indicators. The Agent Readiness score reflects how far your website is in supporting AI agent interaction.
Frequently asked questions
Which protocol should I implement first?
Start with the basics: ensure your robots.txt — bibliotheekterm correctly addresses AI bots and that you have an llms.txt — bibliotheekterm. The next step is MCP if you have an API or service that AI agents would want to use. OAuth Discovery is relevant if you want to expose secured functionality. A2A is currently most relevant for platforms wanting to support multi-agent workflows.
Are these protocols stable enough for production?
MCP is the most mature and is actively used in production environments. A2A is growing rapidly but is younger. Web Bot Auth is still in the specification phase. It's wise to implement MCP first and follow the other protocols as they mature.
What if a new protocol comes along that replaces these?
Complete replacement is unlikely. The protocols serve different use cases and will likely coexist, similar to how HTTP, WebSocket, and gRPC coexist. By investing in agent readiness now, you build experience and infrastructure that remains valuable with future protocols as well.
Does my SMB website need agent protocols?
For most SMB websites, the basic steps (llms.txt, robots.txt AI rules, structured data — bibliotheekterm) are sufficient. MCP and A2A become relevant if you have a service or platform with which AI agents can meaningfully interact. Focus on the content layer (AEO — bibliotheekterm) first and build the protocol layer afterward.
How does this relate to existing API standards like REST and GraphQL?
REST and GraphQL are designed for developer-to-machine communication: a developer writes code that calls specific endpoints. Agent protocols are designed for machine-to-machine communication: an AI agent discovers what's available on its own and determines which actions are needed. They coexist, with agent protocols often forming a layer on top of existing APIs.
