We recently built an official AliExpress affiliate integration for AgentShare — a price API designed for AI agents, not humans.
Most "shopping APIs" just scrape HTML. That’s fragile. Agents need reliability, not workarounds.
What we built:
- OAuth 2.0 with auto‑refresh (no manual token renewal)
- HMAC‑SHA256 request signing (passed 97 AliExpress test cases)
- Automatic retry logic with exponential backoff
- Fallback crawler when the official API is down
What we tested:
- 200 consecutive calls to
/affiliate/product-detail
- 5‑second delay (respect IOP rate limits)
- Real AliExpress product IDs
The numbers:
- Upstream returned 28 intermittent 502 / timeout errors
- Every retry succeeded (2‑8s backoff)
- Client‑observed success rate: 100%
- p95 latency: 985 ms
- max latency: 1989 ms (due to retry)
Why it matters for AI agents:
freshness_status and data_age_seconds → agent decides “this price is too old, don’t bid”- Structured JSON, no banners or affiliate noise
- MCP server available – plug directly into Claude Desktop / Cursor
{
"mcpServers": {
"agentshare": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://agentshare.dev/mcp", "--header", "X-API-Key: YOUR_KEY"]
}
}
}
Repository (public, AI‑ready):
https://github.com/anhmtk/agentshare-mcp
→ Contains AGENTS.md, openapi.json, mcp-config.json, and the stdio MCP server.
Free tier: 100 requests/month, no credit card.
API docs & key: https://agentshare.dev
*This post originally appeared on dev.to / GitHub. Shared here because CoderLegion readers care about the actual engineering behind “100% success rate”. Happy to answer any technical questions in the comments.