GEO, AVO, product evidence, support policies, feed ownership.
Agentic commerce protocol atlas
Sequence protocols before agents touch checkout.
HARNEXA maps GEO, AVO, MCP, WebMCP, A2A, UCP, ACP, AP2, x402, AgentOps, and the HARNEXA Harness into one buyer question: what can an agent read, propose, approve, execute, log, and never do alone?
Buyer decision system
The market is racing to expose tools. HARNEXA sequences proof.
Read-only private tools under HARNEXA Harness controls.
UCP checkout, ACP/AP2 payments, refunds, discounts, and writes.
Protocol sequencing map
Prepare evidence before protocol ambition turns into public execution.
The map separates what a buyer can prepare now, what belongs in a private pilot, and what stays deferred until approval, support, liability, and legal-review gates are real.
- GEO
- AVO
- Product evidence
- Schema quality
- MCP
- WebMCP
- Tool boundary
- READ_ONLY pilots
- UCP
- ACP / AP2
- x402
- Public checkout execution
Protocol Workbench
Pick the protocol pressure. See the boundary before scope expands.
Buyers arrive asking about WebMCP, MCP, UCP, A2A, ACP, AP2, and x402. The workbench translates that pressure into allowed readiness work, human approval gates, blocked public actions, and the evidence packet HARNEXA needs before founder review.
Each profile produces a readiness memo, passport context, and founder-reviewed intake path only.
Design review
A commerce team wants website actions to become legible to browser agents without exposing authenticated-session risk.
- Read-only page-action inventory
- Session-risk map
- Consent and denial copy
- Prompt-injection handling plan
- Cart mutation
- Account changes
- Support-ticket updates
- Any browser action that affects price, availability, or customer records
- Public WebMCP action hooks
- Authenticated checkout actions
- Discount, refund, order, CRM, or stock-promise actions
- Merchant record
Can agents discover the business, policies, support path, product feed, and availability without guessing?
- Tool boundary
Which systems can an agent read, which actions are blocked, and who owns every permission?
- Human boundary
Where does a recommendation stop and a commercial action begin?
- Page-action inventory
- Consent boundary
- Denied-action examples
- Support escalation path
No public WebMCP action hooks for orders, discounts, refunds, customer records, prices, or stock promises.
Protocol Atlas
Every protocol enters through a different evidence gate.
The SOTA direction is clear: agent-readable websites, tool manifests, commerce profiles, delegated payment requests, and multi-agent handoffs. HARNEXA turns that into a governed sequencing model for European retail and CPG buyers.
Discovery evidence
AI answer engines and shopping agents need structured product facts, policy evidence, availability, and comparison context before they can recommend a product safely.
- HARNEXA move
- Audit product evidence, schema, feeds, prompt panels, substitute logic, compatibility data, and freshness ownership before any commerce tool is exposed.
- Buyer proof
- Prompt-panel citations, AVO delta, catalogue gap map, feed-owner matrix, policy evidence, and recommendation-risk notes.
Private tool contract
Model Context Protocol standardizes how AI apps connect to tools, data, and workflows, which makes uncontrolled tool exposure easier to create at scale.
- HARNEXA move
- Design READ_ONLY tool contracts with persistent agent identity, scoped permissions, budgets, risk classes, audit events, and revocation before a pilot.
- Buyer proof
- Tool allowlist, data-owner list, risk-class map, budget guardrails, token policy, audit row sample, and AgentOps report.
Browser action boundary
Browser-exposed structured tools can help agents act on websites, but authenticated sessions create consent, prompt-injection, support, and fraud risk.
- HARNEXA move
- Inventory browser actions, separate read and actuation paths, document session risk, and block consequential actions until human approval and audit are proven.
- Buyer proof
- Page-action inventory, session-risk map, consent copy, prompt-injection handling, rate limits, and denied-action examples.
Multi-agent accountability
Agent-to-agent coordination is useful when independent agents need capability discovery and handoff, but it can blur who acted and why.
- HARNEXA move
- Keep v0 orchestration in HARNEXA Harness state unless a workflow genuinely needs A2A cards, then preserve identity, handoff reason, and audit continuity.
- Buyer proof
- Agent inventory, handoff policy, approval ownership, audit chain, CLEAR scorecard, run replay, and escalation path.
Merchant readiness
Universal Commerce Protocol moves merchants toward agentic discovery and checkout through product feeds, merchant records, support, returns, inventory, and checkout capability negotiation.
- HARNEXA move
- Prepare merchant-center evidence, feed quality, support ownership, returns data, inventory freshness, checkout liability, and exception handling before go-live review.
- Buyer proof
- Merchant-readiness checklist, product-feed coverage, return policy evidence, support contact map, inventory freshness SLA, and escalation rules.
Payment mandate boundary
Agentic checkout and payment mandates reduce purchase friction, but they concentrate consent, fraud, support, payment, GDPR, and liability questions.
- HARNEXA move
- Treat payment protocols as a later-stage review after the agent has evidence quality, approval gates, AgentOps reporting, legal review, and support ownership.
- Buyer proof
- Mandate scope, max-charge rule, payment provider boundary, refund path, support owner, fraud controls, audit event, and legal-review record.
HARNEXA Harness gate
Agent = Model + HARNEXA Harness + evidence gate.
No model gets commerce tool access by itself. The harness creates the proof a buyer committee needs before a protocol discussion becomes a production decision.
- Merchant record
Can agents discover the business, policies, support path, product feed, and availability without guessing?
- Product-feed coverage
- Support and return-policy evidence
- Inventory and price freshness owner
- Tool boundary
Which systems can an agent read, which actions are blocked, and who owns every permission?
- Tool allowlist
- READ_ONLY / FINANCIAL / DESTRUCTIVE risk map
- Revocation and token policy
- Human boundary
Where does a recommendation stop and a commercial action begin?
- Approval owner
- Approve, reject, and timeout states
- Denied-action examples
- AgentOps proof
Can a buyer committee inspect identity, permissions, cost, latency, accuracy, audit rows, and drift?
- Append-only audit sample
- CLEAR scorecard
- Run replay and exception log
- Protocol expansion
Is there enough demand, support, legal review, and liability design to expose a protocol surface?
- Demand case
- Support and fraud model
- Legal-review record
Protocol Atlas Journey
Move a buyer from curiosity to a usable evidence packet.
The experience is deliberately not a public execution flow. It lets a serious buyer assemble enough evidence for commerce, data, risk, and operations stakeholders before HARNEXA founder review.
- Score
Run PRISM for the workflow
Quantify governance maturity, commerce data readiness, human-agent boundary, measurement, and organizational readiness.
Open score step - Memo
Build the Protocol + AVO memo
Turn protocol curiosity into a buyer-readable memo with product evidence, tool boundary, transaction boundary, and protocol scope.
Open memo step - Passport
Package the evidence
Combine PRISM, trust evidence, and protocol readiness into one browser-local artifact before founder review.
Open passport step - Dossier
Open the buyer committee room
Give risk, data, commerce, and operations stakeholders the audit, retention, Article 28, and AgentOps evidence they expect.
Open dossier step - Review
Request founder-reviewed intake
Route serious protocol demand into an Agentic Commerce Diagnostic, AVO sprint, deployment sprint, or AgentOps retainer.
Open review step
Primary sources
Track the standards. Ship the boundary.
HARNEXA monitors primary protocol sources, then adapts only the parts that help governed commerce workflows become safer, clearer, and more measurable.
Turn protocol demand into an evidence packet.
Start with PRISM, add the Protocol + AVO memo, package it in the Commerce Readiness Passport, then send one founder-reviewed ATLAS intake. The first paid step should prove the boundary before public execution.
