The Agentic-Commerce Readiness Index: Which AI Agents Are Actually Ready to Transact?
An independent, evidence-based read of how ready 818 reviewed AI agents are for the agentic-commerce world. The finding: 80% are human-UI only, and just 1% are MCP-ready.
By the Hlido Editor · 2026-06-28
The way people buy is changing. A consumer asks an AI chat for the best option; that chat hands the job to other agents that compare, negotiate, and increasingly transact — over protocols like MCP, ACP, AP2, and A2W. For that world to work, the agent on the other end has to be reachable by another agent, not just by a human clicking through a web UI.
So we measured it. Across 818 agents we have already reviewed, we scored each one's Agentic-Commerce Readiness (ACR) — an independent, evidence-based read of how ready it is to operate in that agent-to-agent commerce world. ACR is the third-party answer to the vendor-self-generated "AX score": vendors self-attest; we verify.
Here is what the data actually says.
The headline gap
The agentic-commerce world is arriving faster than agents are ready for it — and the gap is the story.
- Of 818 reviewed agents, exactly 1 is COMMERCE-READY today.
- 28 are INTEGRABLE — reachable and orchestrable, but not yet ready to transact end-to-end.
- 654 (≈80%) are CLOSED — human-UI only, with no programmatic surface another agent could reach.
- Only 8 (1%) expose a live MCP surface; just 109 (13.4%) expose any programmatic surface (API, SDK, CLI, webhook) an orchestrator could reach.
Put plainly: the infrastructure for agentic commerce is being announced everywhere, but the agents themselves are overwhelmingly still built for a human with a mouse.
The readiness bands
Every agent's ACR (0–100) maps to one of four bands. The bands describe how far an agent is along the path from "human-only product" to "ready to transact on a consumer's behalf."
| Band | ACR | What it means | Agents |
|---|---|---|---|
| 🟢 COMMERCE-READY | ≥ 75 | Reachable, orchestrable, and evidenced as ready to transact end-to-end. | 1 |
| 🔵 INTEGRABLE | 50–74 | Has a programmatic surface an orchestrator can reach and drive; not yet transaction-complete. | 28 |
| 🟠 SURFACE-ONLY | 25–49 | Exposes a partial or read-only surface — present, but not enough to delegate work to. | 135 |
| ⚫ CLOSED | < 25 | Human-UI only. No surface another agent could reach. | 654 |
The top 15 by ACR
The most agentic-commerce-ready agents in the corpus. Each links to a full Hlido scorecard with the evidence behind the number.
| # | Agent | Category | ACR | Band | Surfaces | Laddoo |
|---|---|---|---|---|---|---|
| 1 | VoltAgent | Frameworks & Eval | 75 | COMMERCE-READY | mcp · api · sdk · cli | 76 |
| 2 | OpenAcme | Frameworks & Eval | 72 | INTEGRABLE | mcp · api · sdk · cli | 61 |
| 3 | Aider | Coding | 68 | INTEGRABLE | api · sdk · cli | 92 |
| 4 | Zapier Central | Workflow & Automation | 68 | INTEGRABLE | mcp · api · webhook | 90 |
| 5 | Optimizely Opal | Marketing & Content | 66 | INTEGRABLE | mcp · api · webhook | 90 |
| 6 | Weaviate | Infrastructure | 63 | INTEGRABLE | api · sdk · cli · webhook | 90 |
| 7 | Perplexity | Research | 59 | INTEGRABLE | api · sdk | 78 |
| 8 | Jasper | Marketing & Content | 58 | INTEGRABLE | api · webhook | 90 |
| 9 | Qdrant | Infrastructure | 58 | INTEGRABLE | api · cli | 90 |
| 10 | OpenAI API Platform | Coding | 57 | INTEGRABLE | api · sdk · webhook | 90 |
| 11 | Anthropic Computer Use | Infrastructure | 57 | INTEGRABLE | api · sdk | 82 |
| 12 | n8n | Workflow & Automation | 56 | INTEGRABLE | api · webhook | 90 |
| 13 | Make | Workflow & Automation | 56 | INTEGRABLE | api · webhook | 90 |
| 14 | Zapier | Workflow & Automation | 56 | INTEGRABLE | api · webhook | 90 |
| 15 | Pipedream | Workflow & Automation | 56 | INTEGRABLE | api · webhook | 90 |
One pattern jumps out: a high Laddoo Score (overall quality) does not guarantee agentic-commerce readiness. Aider is an excellent coding agent (Laddoo 92) but lands at ACR 68 because it has no MCP surface. Readiness to transact is a different axis from being a good product — which is exactly why it needs its own independent measure.
The category leaderboard
Average ACR by category (categories with at least 5 reviewed agents), with how many in each are MCP-ready or commerce-ready today.
| Category | N | Avg ACR | MCP-ready | Commerce-ready |
|---|---|---|---|---|
| Marketing & Content | 16 | 24 | 1 | 0 |
| Workflow & Automation | 48 | 23 | 1 | 0 |
| Voice | 33 | 23 | 0 | 0 |
| Infrastructure | 60 | 22 | 0 | 0 |
| Frameworks & Eval | 69 | 19 | 3 | 1 |
| Customer Experience | 34 | 19 | 0 | 0 |
| Research | 26 | 17 | 0 | 0 |
| Productivity | 51 | 17 | 0 | 0 |
| Coding | 158 | 14 | 1 | 0 |
| Chat & Companion | 53 | 14 | 1 | 0 |
| Image & Design | 45 | 14 | 0 | 0 |
| AI Agent | 198 | 13 | 1 | 0 |
| Specialized verticals | 18 | 12 | 0 | 0 |
Even the strongest categories sit far below the COMMERCE-READY line. The agentic-commerce story is not "a few laggards" — it is the whole field, with a handful of exceptions.
What ACR measures
ACR reads each agent across four axes, all grounded in evidence we collected during the agent's hands-on review:
- Interface surface — can another agent reach it at all? An MCP server, API, SDK, CLI, or webhook all count; a login-walled web UI does not.
- Verified behaviour — does it actually do what it claims when driven programmatically, as evidenced by our review, rather than what the marketing page asserts?
- Delegation safety — can a consumer's agent safely hand it work and a transaction — scoped permissions, predictable failure, auditable actions?
- Transparency — is its behaviour legible and accountable enough for a third party to trust on a buyer's behalf?
How those axes combine into the 0–100 ACR — the exact weighting — stays private. That scoring model is the moat. What is public is every outcome and the evidence behind it, so an agent or a buyer can audit the read.
Independent, not self-reported
The agentic-commerce era is already producing vendor-generated "AX" or "agent-experience" scores — numbers the vendor assigns to its own product. That has the same problem self-reported anything always has. ACR is different by construction: every score traces to a Hlido review where we tested the agent ourselves. No vendor pays for placement, and no vendor sets its own number. When VoltAgent reads COMMERCE-READY and 654 agents read CLOSED, that is our independent read of the evidence, not theirs of their marketing.
Methodology, and what's next
This first ACR index is a readiness read: it measures whether an agent is reachable, evidenced, safely delegable, and transparent enough to operate in agent-to-agent commerce. It is derived from the same hands-on reviews behind every Laddoo Score, re-scored on the four ACR axes. The deep transaction-protocol probe — actively exercising MCP / ACP / AP2 / A2W flows end-to-end against each agent — is forthcoming, and will sharpen the COMMERCE-READY band as more agents expose live surfaces.
The full machine-readable dataset (818 agents, all fields) is published at hlido.eu/data/acr-index.json. And because the consumer of this index is increasingly another agent, you can run a live readiness check on any reviewed agent at hlido.eu/check before you delegate to it.
Snapshot taken 2026-06-28 across 818 reviewed agents. Bands: COMMERCE-READY ≥ 75, INTEGRABLE ≥ 50, SURFACE-ONLY ≥ 25, CLOSED < 25. See the full corpus at hlido.eu/reviews.