FCHubFCHub.co

Usage & Examples

How to talk to your FluentCart store through AI. I show you what to ask, how to ask it, and what to expect back.

You're connected. Now what?

You just... talk. No special syntax. No commands to memorise. No tool names to remember. I built 194 tools with descriptions so rich that any halfway decent AI can figure out which one to use from plain English. Describe what you want and let the AI sort it out.

How to Talk to It

You don't need to know my tool names or API parameters. The AI translates your intent into the right calls:

Don't say thisSay this instead
"Call fluentcart_order_list with per_page=10""Show me the latest 10 orders"
"Use fluentcart_product_create with type digital""Create a new digital product called Pro License"
"Execute fluentcart_coupon_create""Make a 20% off coupon code"

The AI picks the right tool, fills in the parameters, and calls the MCP server. You'll see it working — most clients show which tools are being called in real time.

Specificity Wins

"Show me orders" works. "Show me pending orders from the last 7 days sorted by total, highest first" works better. The more context you give, the more precise the results. I designed every tool schema to accept useful filters — help the AI use them.

Example Prompts by Category

I've organised these by the things people actually ask. Pick a section, steal a prompt, adjust to taste.

Tips I've Learned

Give it IDs when you have them

The AI can't guess which order you mean. "Show me order #1234" is instant. "Show me that order from yesterday" requires the AI to search, filter, and hope for the best.

Chain requests

You can absolutely ask for multiple things at once:

  • "Find customer jane@example.com, show her order history, and create a 10% loyalty coupon for her"

The AI will call multiple tools in sequence. It's surprisingly good at this.

Ask for analysis, not just data

My tools return raw data, but the AI can interpret it:

  • "Summarise this week's sales performance"
  • "Compare this month's revenue to last month and tell me if we're on track"
  • "Which of my top 10 products had declining sales this month?"

I provide the data. The AI provides the insight. Division of labour.

Describe support scenarios

Rather than issuing individual commands, describe the whole situation:

  • "Customer john@example.com says he was charged twice. Check his recent orders and see if there's a duplicate."
  • "Customer wants to upgrade from Basic to Pro. What subscriptions do they have and what are the options?"

The AI will investigate, call the right tools, and report back.

What It Can't Do

I built a lot of tools, but I'm not delusional. There are limits:

Can't DoWhy
Charge credit cardsPayments require customer interaction (redirects, card forms). The MCP server can mark orders as paid, but it can't process payments.
Send emailsEmail templates live in FluentCart's UI. Not an API surface I can reach.
Edit checkout pagesThat's frontend work — HTML/CSS/JS. Not API territory.
Upload product imagesFile uploads aren't part of the current API surface. Use WordPress admin for media.
Configure shipping zonesOne-time setup work. I'm adding this in v1.1.
Manage tax ratesComplex and legally sensitive. Coming in v1.1.
Touch other WordPress dataThe MCP server only talks to FluentCart endpoints. It can't read your blog posts, other plugins, or WordPress users.

A Note About Money

FluentCart stores all monetary values in the smallest currency unit. Cents for USD/EUR, grosze for PLN, pence for GBP. This means:

  • $49.99 is stored as 4999
  • €100.00 is stored as 10000
  • 40.00 PLN is stored as 4000

The AI usually converts these when presenting results to you. But if you see numbers that look 100x too large, that's why. When creating products or coupons, just say the human amount — "price it at $49" — and the AI will handle the conversion. I documented the currency format in every tool description precisely so the AI gets this right.

Dynamic Mode

By default, all 194 tools are registered as static MCP tool definitions. That's ~20K tokens of context the LLM loads on every turn. If your context window budget matters (and it should), start the server with --mode dynamic:

npx fluentcart-mcp --mode dynamic

Or in your Claude Desktop config:

{
  "mcpServers": {
    "fluentcart": {
      "command": "npx",
      "args": ["-y", "fluentcart-mcp", "--mode", "dynamic"],
      "env": { ... }
    }
  }
}

Instead of 194 tool definitions, the AI gets 3:

  1. fluentcart_search_tools — Find tools by keyword and category
  2. fluentcart_describe_tools — Get full schemas for specific tools
  3. fluentcart_execute_tool — Run any tool by name

The AI searches for what it needs, reads the schema, then executes. Same tools, same results — just discovered lazily instead of dumped upfront. ~96% fewer tokens in the initial context.

When to Use Dynamic Mode

If you have a large conversation or use other MCP servers alongside this one, dynamic mode keeps your token budget sane. If you only use FluentCart and value speed over savings, static mode is fine.

MCP Resources

Four read-only resources for reference data. These are browsable context the AI can read without calling tools — useful for grounding conversations with store configuration:

Resource URIDescription
fluentcart://store/configStore settings and configuration (currency, locale, modules)
fluentcart://store/countriesSupported countries and their details
fluentcart://store/payment-methodsConfigured payment methods
fluentcart://store/filter-optionsAvailable filter options for orders, products, customers

Resources are application-controlled — the client decides when to read them. Static data is cached in-memory, so repeated reads don't hit the API.

MCP Prompts

Five pre-built workflows that guide the AI through multi-step operations. Think of them as recipes — the AI knows which tools to call and in what order:

PromptArgumentsWhat It Does
analyze-store-performancestartDate, endDateRevenue overview, KPIs, and best sellers for a date range
investigate-orderorder_idDeep-dive into order details, transactions, and activity timeline
customer-overviewcustomer_idFull customer profile with stats, addresses, and spending history
catalog-summary(none)Catalog health report — product count, top sellers, store metrics
subscription-healthstartDate, endDateSubscription churn, renewal success, and revenue forecast

Not all MCP clients support prompts yet. Claude Desktop and Claude Code do. If yours doesn't, just ask for the same thing in plain English — the AI will figure out the same tool sequence on its own.

Response Caching

Static reference data is cached in-memory so the AI doesn't re-fetch the same 250-country list every time it needs to validate an address:

EndpointCache TTL
Countries list1 hour
Filter options10 minutes
App init (store config)10 minutes
Report metadata10 minutes

The cache lives in the server process. Restart the server to clear it. No external dependencies, no Redis, no drama.

Next Steps

On this page