Agent builder quickstart
Value: Give Claude, ChatGPT, Gemini — or any MCP-aware agent — safe, task-shaped tools that turn natural language into real Solid Commerce work: list products, check pending shipments, run sales reports, reprice listings. Reads run instantly; writes always confirm first.
Choose this if: You want an AI assistant to operate your store, not just answer questions about it.
Two ways to connect
| | Local (stdio) | Hosted (remote) |
|---|---|---|
| Best for | Claude Desktop, Claude Code, Codex / Gemini / Grok CLI, custom agents | Claude.ai, ChatGPT, Gemini app |
| Runs | on your machine (npx) | our servers |
| Auth | an API key you mint once | OAuth — you just log in |
| Endpoint | npx -y @solidcommerce/mcp-server@alpha | https://mcp.solidcommerce.com/mcp |
Local gives you a key you control and works everywhere today. Hosted needs no install at all — each teammate logs in with their own Solid Commerce account.
Step 1 — Mint an API key (local path only)
-
Sign in to your workspace and open Developer → API keys → New (the
/developer/keyspage). -
Pick the smallest scope set your agent needs. A good starting set for a read-mostly assistant:
catalog:products:read orders:orders:read orders:shipments:read reports:sales:read reports:shipments:read inventory:inventory_items:read -
Copy the key once — the secret is never shown again. It looks like
ab12cd34_R8fK…(an 8-character prefix, an underscore, then the secret).
Lost a key? Revoke it and mint a new one — rotation is one click on the same page, and the audit log shows every call a key made.
Step 2 — Connect your client
Claude Desktop
Edit the Claude Desktop config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"solidcommerce": {
"command": "npx",
"args": ["-y", "@solidcommerce/mcp-server@alpha"],
"env": { "SOLIDCOMMERCE_API_KEY": "YOUR_KEY" }
}
}
}
Quit and reopen Claude Desktop (MCP servers load at startup), then ask:
List 5 products from my Solid Commerce catalog.
The server targets the Solid Commerce platform API out of the box; set
SOLIDCOMMERCE_API_BASE_URL in env only if you need to point elsewhere.
Claude Code
Two commands inside Claude Code — it prompts for your key on install:
/plugin marketplace add https://dev.azure.com/takeoffcommerce/SC-Base/_git/claude-code-plugin
/plugin install solidcommerce@solidcommerce
Full guide: Claude Code.
ChatGPT (OpenAI desktop & web)
ChatGPT connects to the hosted endpoint — no install, no key:
- Open Settings → Connectors (custom connectors require a paid ChatGPT plan; menu naming varies slightly by release — check OpenAI's current docs if it moved).
- Add a connector with the MCP server URL
https://mcp.solidcommerce.com/mcp. - Sign in when prompted — the Solid Commerce consent screen lists the requested scopes in plain English, with a company picker if you belong to more than one company.
Then ask: "How many orders came in over the weekend?" More: ChatGPT.
Gemini
Gemini CLI (works today, local): add to ~/.gemini/settings.json:
{
"mcpServers": {
"solidcommerce": {
"command": "npx",
"args": ["-y", "@solidcommerce/mcp-server@alpha"],
"env": { "SOLIDCOMMERCE_API_KEY": "YOUR_KEY" }
}
}
}
Run gemini, confirm /mcp shows solidcommerce connected, and ask the
same catalog question. Full guide: Gemini CLI.
Gemini app (gemini.google.com): consumer connectors are rolled out through Google's extension program; our hosted endpoint is ready for it. Track status on Gemini — and use the CLI in the meantime.
Step 3 — What your agent can do
The server exposes 96 task-shaped tools spanning every domain of your store — catalog, orders, reports, listings, inventory, vendors, and platform administration. Real examples:
| Ask | Tool the agent uses |
|---|---|
| "List my products" | catalog_products_list |
| "How many orders are pending shipment?" | reports_pending_shipments_list |
| "What shipped this week?" | orders_shipments_list |
| "Top sellers this month?" | reports_top_sellers_list |
| "What's low on stock?" | reports_inventory_low_stock_list |
| "Fulfill order 1234" | fulfill_order (asks you to confirm first) |
Two guarantees hold everywhere:
- Reads vs writes. Every tool is annotated read-only or destructive. Destructive tools (fulfill, cancel, reprice, reply) surface a confirmation with the exact action and target before running.
- Scopes are enforced server-side. A tool outside your key's scopes fails
with a clean
403naming the missing scope — the agent cannot escalate itself. Full list: MCP tool reference.
Step 4 — Use it from a custom agent
Any MCP client can spawn the same server. With the Anthropic Agent SDK:
import { ClaudeAgent } from '@anthropic-ai/claude-agent-sdk';
const agent = new ClaudeAgent({
mcpServers: {
solidcommerce: {
command: 'npx',
args: ['-y', '@solidcommerce/mcp-server@alpha'],
env: { SOLIDCOMMERCE_API_KEY: process.env.SOLIDCOMMERCE_API_KEY },
},
},
});
await agent.run('How many orders are pending shipment?');Troubleshooting
- 401 on every call — the key is wrong, revoked, or expired. Mint a fresh
one at
/developer/keysand update your config. - 403
insufficient_scope— the error names the exact scope to add. Rotate the key with the extra scope (or mint a new one). - Tools don't appear — MCP servers load when the client starts. Fully restart Claude Desktop / your CLI after config changes.
- Sanity check — ask "How many orders are pending shipment?" It exercises auth, scopes, and the reports surface in one question.
Next steps
- MCP tool reference — every tool, every parameter.
- Customer admin quickstart — script the same API directly.
- All AI surfaces — Claude, ChatGPT, Gemini, Copilot, and the CLIs.