Bring orders, inventory, integrations, and business rules into your agent’s workflow. StateSet MCP servers let compatible AI clients discover tools, read operational context, and carry out permitted actions.
Setup examples for Cursor, Claude Desktop, and Claude Code.
YOU ASK
“Which orders need attention, and what should we do next?”
Your AI client
Understands the request and selects available tools.
StateSet MCP server
Retrieves records from the connected system.
A grounded next step
Returns findings and proposes actions for review.
Illustrative flow. The tools and data available come from the server and account you connect.
Each server connects to a different part of your operations. Choose one to see its requirements, connection example, and a first prompt.
Local commerce engine
Explore commerce data and build workflows around a local SQLite database. A useful starting point for developers and product teams.
Inspect orders and stock, prototype a commerce assistant, or preview an operational change.
StateSet API access
Connect an assistant to the StateSet commerce API for orders, returns, inventory, manufacturing, fulfillment, and finance.
Investigate an order, review a return, or bring stock and fulfillment context into a support workflow.
Connected systems
Give agents visibility into the jobs and integrations that move orders, products, inventory, and returns between systems.
Understand failed sync jobs, inspect integration health, and prepare a targeted recovery plan.
Trading partner operations
Work with trading partner documents, validation results, delivery evidence, acknowledgments, and reconciliation through the EDI gateway.
Trace a purchase order through shipment and invoice documents, or investigate an acknowledgment gap.
Voice operations
Inspect voice agents, number routing, knowledge sources, and call results through the StateSet Voice API.
Review call outcomes, diagnose routing, or find gaps in the knowledge available to a voice agent.
Policy reasoning
Evaluate proposed decisions against configured rules and supplied facts, with a verdict and supporting reasoning.
Check refund eligibility, identify missing facts, and review policy reasoning before an agent acts.
Start with a test environment and a small set of tools. These examples run local servers over stdio; your client starts and manages the process.
Local · stdio
Node.js 20.20+ and an absolute path to a local commerce database. No StateSet API key is needed for the local engine.
The client launches the CLI through npx. Replace the database path with your own. A new database starts empty; use an existing test database for meaningful results.
Merge this entry into your personal MCP configuration. Reload the client and check that the server is enabled in MCP settings.
Check that the client reports a connected server and lists its tools. Ask which read tools it can use before trying the example below.
Replace every YOUR_ placeholder and absolute path before connecting. Keep credentials in your client’s private configuration or secret store, outside shared repositories.
{
"mcpServers": {
"stateset-cli": {
"command": "npx",
"args": [
"-y",
"-p",
"@stateset/cli",
"stateset-mcp",
"--db",
"/absolute/path/to/store.db",
"--profile",
"core"
]
}
}
}The core profile keeps discovery focused. Writes are preview-only by default; enabling execution requires operator configuration and the appropriate kernel policy.
List up to five orders in this database and summarize their status. Use read tools only. If there are no orders, tell me the database is empty.What to expect: A short status report based on your local data, with order references you can inspect.
Choose another serverYour client asks the server which tools are available and what inputs they accept.
The agent calls a tool to retrieve the orders, stock, documents, or policies relevant to your request.
You inspect the proposed action, its target, and the permissions required before approving a write.
The agent checks the tool result and reads the updated record to confirm what actually happened.
Name the order, SKU, tenant, or date range. Ask for record IDs and evidence so you can check the result.
Start with investigation. For changes, review the exact records and inputs, then use client approvals and server permissions.
Ask the agent to read back the final state. If a write times out, verify whether it completed before retrying.
Use a remote MCP endpoint that your client can reach. Local commands and backend REST URLs cannot be substituted for a hosted MCP connection.
For Sync and EDI, obtain the MCP endpoint and authentication requirements from your deployment administrator. Confirm your client supports that transport and sign-in method.
NSR hosted connection
https://nsr.stateset.com/mcpNo. Start with the server for your workflow. Add another when you need its capabilities. Available tools depend on the server version, profile, tenant, and permissions.
With stdio, your client starts a program on your machine and communicates with it directly. A remote MCP client connects to a deployed HTTP endpoint. A REST API URL is not automatically an MCP endpoint; use the MCP URL and authentication supplied for your deployment.
A cloud-hosted assistant cannot launch a program on your laptop through a stdio configuration. Use a supported remote MCP endpoint and complete its authentication flow. The NSR hosted connection is described below; Sync and EDI depend on your deployment.
Check the selected account, tenant, database path, and tool permissions. A new local iCommerce database is empty. A connection confirms the protocol is working, not that it contains your production data.
For authentication errors, check the key, tenant, and endpoint. For a missing executable, check Node.js and use absolute paths. For missing tools, check your profile and server version. For a timeout after a write, inspect the target record before retrying.
A prompt describes your intent; it does not change server permissions. Use scoped credentials, server policy, and client tool approvals to control execution. The iCommerce CLI examples here leave write execution disabled by default.
Connect a server, explore your data, and turn a focused use case into a working integration.