# Agentic Commerce for Your Team and AI Agents | Kentro

Source: https://kentro.ai/agents/
Last updated: 2026-11-17

Agentic commerce on one system of record: your team works in the app, Kentro AI answers, and your agents connect over MCP and the API with the access you grant.

## One system of record for your team and your AI agents

Your team, Kentro AI and your own agents work from the same system of record. What each one can do depends on the workflow and the connection, and every agent gets the access you grant and nothing more.

- Works with Claude and Codex
- MCP agents sign in as a user, no key to paste
- Revoke a connection any time
- API and MCP on every Kentro OS plan

- [Connect an agent](https://kentro.ai/agents/mcp/#connect)
- [See what agents can do](https://kentro.ai/agents/capabilities/)

## What is agentic commerce?

Agentic commerce is commerce where AI agents do part of the work: answering questions from live data, preparing documents and flagging what needs a decision, while people approve what matters. Agents need a system of record they can read and act on safely. Kentro is a commerce suite built to be that system for operations, fulfillment, finance and marketing.

- [Glossary: agentic commerce](https://kentro.ai/glossary/agentic-commerce/)

## Who does what in Kentro

All three work from the same records. What each one can do depends on the workflow and the connection, and the capability matrix lists it.

- **People, in the app**: Screens for the office, the warehouse floor and the handheld scanner. Buying labels, charging a customer’s saved card and warehouse scans stay here. (https://kentro.ai/os/)
- **Kentro AI**: Our assistant. It answers with the permissions of the person asking, builds dashboards, acts on findings and drafts demand events for a person to approve. (https://kentro.ai/agents/kentro-ai/)
- **Your own agents**: Claude, Codex or an agent you build. Over MCP it signs in as a user and reads across the suite. On the Channel API and the Store API it writes core documents with a scoped key. (https://kentro.ai/agents/mcp/)

## Agents get the access you grant, and nothing more

Kentro gives agents no back door. An agent connected over MCP signs in as a user and inherits that user's role, in one workspace. An integration or agent on the API uses a key whose permission groups you choose. Kentro AI works the same way: it sees what the person asking can see. Where an AI output changes something that matters, a person approves it before it takes effect.

- MCP: OAuth sign-in as the user, approved in a browser window
- API: keys with the scopes you choose, read or write
- Data questions run on a read-only database connection
- Every MCP connection is listed in Kentro AI Settings and can be revoked there
- Tenant data is isolated at the database level
- Approvals: Kentro GO drafts and Kentro AI demand events

- [Security at Kentro](https://kentro.ai/security/)

## What agents can do

A dated summary by area: in the app, with Kentro AI, over MCP and over the API. The full matrix lists every workflow.

Strongest access per area, checked on 2026-11-17. Access levels: Read, Write, Write with approval, Not available.

| Area | In the app | Kentro AI | Your agent over MCP | Your agent over API |
| --- | --- | --- | --- | --- |
| Orders | Write | Read | Read | Write |
| Customers | Write | Not available | Not available | Write |
| Inventory | Write | Read | Read | Write |
| Purchasing and planning | Write | Write with approval | Read | Write |
| Fulfillment and shipping | Write | Read | Read | Write |
| Warehouse | Write | Read | Read | Read |
| 3PL billing | Write | Not available | Not available | Not available |
| Finance | Write | Read | Read | Write |
| Analytics | Write | Write | Write | Not available |
| Marketing | Write | Write with approval | Not available | Not available |
| Storefront | Write | Not available | Not available | Write |
| Events and tracking | Not available | Not available | Not available | Write |

- [See every workflow in the full matrix](https://kentro.ai/agents/capabilities/)

## Concrete work, not promises

Four jobs that run on Kentro, each with a person in control.

- **A daily operations digest**: Run the built-in daily-ops-digest prompt from Claude or another MCP client. Kentro AI compares today's orders and revenue with yesterday, checks the fulfillment pipeline and inventory that needs attention, and ends with three recommended actions.
  - Built-in prompt: daily-ops-digest
  - Runs through askCopilot with your permissions
  - Saved in your assistant history like a chat in the app
- **Investigate one order end to end**: Ask about one order and get its status, items, fulfillment orders and warehouse routing, labels and tracking, payment state and channel sync in one answer. The built-in investigate-order prompt asks for exactly that.
  - Built-in prompt: investigate-order
  - Direct tools for quick lookups, askCopilot for the full story
- **Triage reconciliation exceptions**: Ask why an invoice and its QuickBooks record differ. Kentro AI reads both sides, explains the variance and suggests the matching record. It cannot post anything: a person with the right permission accepts the fix in the app. 5-Way Reconciliation comes with every Kentro OS plan. 5-Way Reconciliation
- **Draft marketing for approval**: In Kentro GO, the AI Manager drafts posts, ads, landing pages and emails from your products and orders. Every draft waits in an approvals queue, and nothing is published or sent until a person says yes. Kentro GO comes with every Kentro OS plan. Kentro GO

## The assistants your team already uses

Claude and Codex are supported. Another MCP client that meets the published requirements can use the same URL, and software you build can use the REST APIs.

- **Claude**: Claude on the web and Claude desktop, added as a custom connector. (https://kentro.ai/integrations/claude/)
- **Claude Code and Codex**: One command in the terminal, then a browser sign-in. No token to paste. (https://kentro.ai/agents/mcp/#coding-agents)
- **Other MCP clients**: The same URL works for a client that supports remote Streamable HTTP and OAuth with dynamic client registration and PKCE. (https://kentro.ai/agents/mcp/#other-clients)
- **Your own agent**: Reads over MCP. Writes core documents such as orders, stock and invoices through the Channel API, with a scoped key. (https://kentro.ai/developers/)

## Questions about AI agents in Kentro

### What can an AI agent actually do in Kentro?

Over MCP, an agent reads orders, fulfillment, inventory, reports, findings and reconciliation exceptions with your permissions, and Kentro AI can build dashboards for it. To create or change orders, customers, products, stock levels, shipments, invoices, payments or retailer purchase orders, an agent uses the Channel API with a write-scoped key. Vendor purchase orders, label buying, warehouse scans and settings stay with people. The capability matrix lists every workflow.

### Is it safe to connect an agent to my operations?

An agent gets the access you give it and nothing more. Over MCP it signs in as a user and inherits that user's role, and data questions run on a read-only connection. On the API it uses a key whose scopes you choose. You can revoke an MCP connection in Kentro AI Settings at any time, and admins can see and revoke every connection in the workspace.

### Which assistants work with Kentro?

Kentro supports Claude on the web and Claude desktop, Claude Code and Codex. You add one URL, as a custom connector in Claude or with one command in a coding agent, sign in to Kentro in a browser window and approve the connection. Another MCP client can use the same URL if it meets the published requirements. Agents and integrations you build yourself can use the REST APIs with an API key instead.

### Does agent access cost extra?

API and MCP access are part of every Kentro OS plan, which includes unlimited users and AI agents. Requests that run Kentro AI, such as askCopilot from your own assistant, use the workspace's Kentro AI usage: every Kentro OS plan includes $50 a month, and usage beyond that is metered at published rates up to a cap you set.

### What is the difference between Kentro AI and my own agent?

Kentro AI is the assistant built into Kentro. It knows your workspace, keeps the learnings you give it and runs inside the app. Your own agent is Claude, Codex or software you build, connected over MCP or the API. Over MCP it can call most of the tools Kentro AI uses, or hand a whole task to Kentro AI with askCopilot. Writes to operational records go through the Channel API with a scoped key.

### Can an agent buy shipping labels or change settings?

No. No MCP tool buys labels, adjusts stock or changes settings. The Channel API can record a shipment with its carrier and tracking number, but labels are bought by people in Kentro OS. Over MCP an agent reads, analyzes and prepares work. Software that needs to write orders or stock levels uses the Channel API with a write-scoped key.

## Related

- [Kentro AI, the assistant built into Kentro](https://kentro.ai/agents/kentro-ai/)
- [Kentro MCP server for Claude and Codex](https://kentro.ai/agents/mcp/)
- [Capability matrix: what agents can do](https://kentro.ai/agents/capabilities/)
- [Developers: Channel API, Store API and docs](https://kentro.ai/developers/)
- [Kentro for Claude](https://kentro.ai/integrations/claude/)
- [Glossary: agentic commerce](https://kentro.ai/glossary/agentic-commerce/)
- [Security at Kentro](https://kentro.ai/security/)

## Give your agents a real system to work in

Connect an agent with one URL and one approval, or read the full capability matrix before you connect.

- [Connect an agent](https://kentro.ai/agents/mcp/#connect)
- [See what agents can do](https://kentro.ai/agents/capabilities/)
