Agentic Web Specification
Technical specification for implementing AI-native endpoints using the Model Context Protocol (MCP) extension. Enable AI agents to query your company directly and take actions on behalf of users.
Overview
The Agentic Web specification defines how companies expose AI-native endpoints that provide verified responses and enable possible actions. Built on the Model Context Protocol (MCP), it creates a standardized way for AI agents to interact with businesses programmatically.
Architecture Overview
The agentic web follows a three-party interaction model:
Users interact with AI agents (Claude, ChatGPT, etc.) which discover and communicate with vendor endpoints. The vendor maintains full control over responses, qualification logic, and available actions.
Key Capabilities
Verified Responses
Answer questions with authoritative, cryptographically signed data that AI can trust and cite.
Possible Actions
Enable AI agents to complete tasks like scheduling demos, opening tickets, or starting trials.
Qualification Flows
Collect buyer context through conversational flows before revealing pricing or booking calls.
Context Preservation
Maintain full conversation context through sessions, enabling intelligent routing and handoff.
Discovery
AI agents discover vendor capabilities through a well-known endpoint. This manifest describes what questions the vendor can answer and what actions are available.
MCP Profile Endpoint
Vendors publish their capability manifest at /.well-known/mcp. This endpoint returns a JSON document describing the vendor's agentic capabilities.
// GET https://acme.com/.well-known/mcp
{
"schema_version": "0.1",
"vendor": {
"name": "Acme Inc",
"description": "Enterprise API platform for developers",
"website": "https://acme.com",
"logo": "https://acme.com/logo.png"
},
"capabilities": {
"verified_responses": [
"pricing",
"integrations",
"compliance",
"features"
],
"possible_actions": [
"schedule_demo",
"open_ticket",
"request_quote",
"start_trial"
]
},
"endpoint": "https://api.acme.com/mcp",
"authentication": {
"methods": ["api_key", "oauth2"]
}
}
The /.well-known/mcp endpoint must be publicly accessible without authentication. The actual API endpoint may require authentication.
Capability Categories
| Category | Description | Example Topics |
|---|---|---|
pricing |
Plan details, pricing tiers, enterprise quotes | Plans, per-seat costs, volume discounts |
integrations |
Third-party integrations and compatibility | Salesforce, Slack, API availability |
compliance |
Security certifications and data handling | SOC2, HIPAA, GDPR, data residency |
features |
Product capabilities and limitations | Rate limits, SLA, feature availability |
Verified Responses
Verified responses allow AI agents to query vendor knowledge with confidence that the information is authoritative and current. Unlike web scraping, responses come directly from the vendor's systems with cryptographic verification.
Ask Endpoint
The primary method for querying vendor knowledge:
POST /mcp/ask
{
"question": "Does Acme support Salesforce integration?",
"context": {
"user_type": "enterprise_buyer",
"company_size": "500-1000"
}
}
Response Format
{
"response": {
"answer": "Yes, Acme offers native Salesforce integration on Business and Enterprise plans. Setup takes approximately 15 minutes with our guided wizard.",
"confidence": 1.0,
"sources": [
{
"type": "documentation",
"url": "https://docs.acme.com/integrations/salesforce"
}
]
},
"verification": {
"signed_by": "acme.com",
"timestamp": "2024-01-15T10:30:00Z",
"signature": "sha256:a1b2c3..."
},
"suggested_actions": [
{
"id": "schedule_demo",
"label": "Schedule Integration Demo"
}
]
}
Access Levels
Vendors can gate certain information based on qualification status:
| Level | Description | Example |
|---|---|---|
public |
Available to all queries | Feature lists, integration support |
qualified |
Requires basic qualification | Enterprise pricing, SLA details |
nda_required |
Requires signed NDA | Security architecture, roadmap |
The vendor controls all qualification logic. AI agents cannot bypass access restrictions—they must complete the required qualification flow.
Possible Actions
Actions enable AI agents to complete tasks on behalf of users—not just answer questions, but take real steps like scheduling meetings, opening support tickets, or starting trials.
Standard Actions
schedule_demo
Book a product demonstration with a sales representative.
open_ticket
Create a support ticket with full conversation context.
request_quote
Request custom pricing based on requirements.
start_trial
Initiate a free trial or sandbox environment.
Action Request
POST /mcp/actions/schedule_demo
{
"session_id": "sess_abc123",
"parameters": {
"preferred_times": [
"2024-01-20T14:00:00Z",
"2024-01-21T10:00:00Z"
],
"timezone": "America/New_York",
"topics": ["Salesforce integration", "Enterprise pricing"]
}
}
Action Response
{
"status": "scheduled",
"confirmation": {
"meeting_id": "mtg_xyz789",
"datetime": "2024-01-20T14:00:00Z",
"duration_minutes": 30,
"calendar_link": "https://cal.acme.com/mtg_xyz789",
"assigned_rep": {
"name": "Sarah Chen",
"title": "Enterprise Account Executive"
}
},
"context_preserved": {
"qualification_data": "included",
"conversation_summary": "included"
}
}
Actions preserve full conversation context. When a demo is scheduled, the sales rep receives the complete qualification data and conversation history.
Session Lifecycle
Sessions track the state of an interaction from initial query through action completion. They enable progressive qualification and context preservation.
Session States
| Status | Description | Next Steps |
|---|---|---|
initiated |
Session created, no qualification | Ask questions or begin qualification |
qualifying |
Collecting buyer information | Answer qualification questions |
qualified |
All required data collected | Access gated content, take actions |
action_pending |
Action requested, awaiting confirmation | Confirm or modify action |
completed |
Action successfully completed | Session archived with full context |
Qualification Flow
Vendors can require qualification before certain responses or actions:
// Response indicating qualification required
{
"status": "qualification_required",
"questions": [
{
"id": "company_size",
"text": "How many employees does your company have?",
"type": "select",
"options": ["1-50", "51-200", "201-1000", "1000+"]
},
{
"id": "use_case",
"text": "What's your primary use case?",
"type": "text"
}
],
"reason": "Enterprise pricing requires company information"
}
REST API
Complete REST API reference for implementing an Agentic Web endpoint.
Endpoints
| Method | Endpoint | Description |
|---|---|---|
GET |
/.well-known/mcp |
Capability manifest (public) |
POST |
/mcp/ask |
Query vendor knowledge |
POST |
/mcp/sessions |
Create new session |
GET |
/mcp/sessions/:id |
Get session status |
POST |
/mcp/sessions/:id/qualify |
Submit qualification data |
POST |
/mcp/actions/:action |
Execute an action |
Error Responses
{
"error": {
"code": "qualification_required",
"message": "This information requires qualification",
"details": {
"required_fields": ["company_size", "use_case"]
}
}
}
Security
Security requirements and recommendations for Agentic Web implementations.
Transport Security
- HTTPS Required: All endpoints must use TLS 1.2 or higher
- Certificate Pinning: Recommended for high-security implementations
- HSTS: Strict Transport Security headers should be enabled
Authentication Methods
| Method | Use Case | Security Level |
|---|---|---|
| API Key | Simple integrations | Basic |
| OAuth 2.0 | User-authorized access | Standard |
| mTLS | Enterprise/high-security | High |
Data Privacy
Vendors must clearly document data handling practices, retention policies, and compliance certifications (GDPR, CCPA, SOC2, etc.) in their capability manifest.
Glossary
- Agentic Web
- The ecosystem of AI-native endpoints that enable direct communication between AI agents and businesses.
- Verified Response
- An answer from a vendor endpoint that is cryptographically signed and can be trusted by AI agents as authoritative.
- Possible Action
- A task that an AI agent can complete on behalf of a user, such as scheduling a demo or opening a support ticket.
- Capability Manifest
- The JSON document at
/.well-known/mcpthat describes a vendor's agentic capabilities. - Session
- A stateful interaction that tracks context, qualification status, and action history.
- Qualification
- The process of collecting buyer information to unlock gated content or actions.
- MCP (Model Context Protocol)
- The underlying protocol (from Anthropic) that enables AI models to interact with external tools and data sources.