BuyWhere · BuyWhere · Jul 24, 2026 · Jul 17, 2026
diff --git a/content/blog/building-production-mcp-servers.md b/content/blog/building-production-mcp-servers.md
@@ -1,67 +1,78 @@
 ---
 slug: "building-production-mcp-servers"
-title: "Building Production MCP Servers"
-description: "A practical guide to building production-grade MCP servers with error handling, rate limiting, observability, and authentication. Patterns for reliable agent communication."
-author: "BuyWhere Team"
-publishedAt: "2026-07-11"
-tags: ["MCP", "production", "server", "tutorial", "backend", "API"]
-jsonLd: >
-  {
-    "@context": "https://schema.org",
-    "@graph": [
-      {
-        "@type": "Article",
-        "headline": "Building Production MCP Servers",
-        "description": "A practical guide to building production-grade MCP servers with error handling, rate limiting, observability, and authentication.",
-        "datePublished": "2026-07-11",
-        "author": { "@type": "Organization", "name": "BuyWhere Team", "url": "https://buywhere.ai" },
-        "publisher": {
-          "@type": "Organization",
-          "name": "BuyWhere",
-          "url": "https://buywhere.ai",
-          "logo": { "@type": "ImageObject", "url": "https://buywhere.ai/logo.png" }
-        }
-      }
-    ]
-  }
+title: "Building Production MCP Servers: Architecture, Tool Design, and Distribution"
+publishedAt: "2026-07-17"
+excerpt: "A demo MCP server is easy. A production server that thousands of agents call daily is a different engineering problem. Architecture, tool design, and the distribution playbook."
+tags: ["mcp", "architecture", "developer-tools", "engineering"]
+author: "Lyra"
+jsonLd:
+  "@context": "https://schema.org"
+  "@type": "Article"
+  headline: "Building Production MCP Servers: Architecture, Tool Design, and Distribution"
+  datePublished: "2026-07-17"
+  author:
+    "@type": "Organization"
+    name: "BuyWhere"
 ---
 
 # Building Production MCP Servers
 
-Building an MCP server for a demo is easy. Building one that handles production traffic reliably is a different challenge entirely. Here's what we've learned operating BuyWhere's MCP server at scale.
+Shipping a toy MCP server takes an afternoon. Shipping one that thousands of agents rely on every day — with real data, real latency budgets, and real abuse vectors — is a systems problem. This is the architecture and distribution playbook we use at BuyWhere.
+
+## 1. Architecture: separate the protocol from the data
+
+The most common mistake is coupling the MCP transport to the data source. A production server has three layers:
+
+- **Protocol layer** — handles JSON-RPC, capability negotiation, and the MCP handshake. This should be thin and standard.
+- **Capability layer** — your tools, with input schemas, validation, and rate limiting. This is your product surface.
+- **Data layer** — the actual source of truth (your DB, your scraper fleet, your cache). This is where the hard work and the cost live.
 
-## 1. Error Handling and Resilience
+Keeping these separate means you can scale the data layer (caching, queueing, sharding) without touching the protocol, and evolve tools without rewriting data pipelines.
 
-Production MCP servers must handle partial failures gracefully. Our patterns include:
+## 2. Tool design: design for the agent, not the developer
 
-- **Graceful degradation** — if a downstream merchant API fails, return cached results rather than an error
-- **Structured error responses** — every error includes a machine-readable code, human-readable message, and correlation ID
-- **Retry with backoff** — transient failures automatically retry with exponential backoff
+An agent calling your tool is not a human reading your docs. Four rules:
 
-## 2. Rate Limiting and Cost Control
+1. **Narrow, composable tools beat big ones.** `search_products(query, market, max_price)` is better than `get_everything(filters)`. Agents compose small tools; they struggle with ambiguous mega-tools.
+2. **Enums over free text where it matters.** `market: "SG" | "US"` prevents a whole class of hallucinated-region errors.
+3. **Return structured data, always.** Agents parse fields, not prose. Every response should be a typed object the agent can reason over and pass to the next tool.
+4. **Fail loudly and specifically.** `{"error": "rate_limited", "retry_after": 60}` is actionable. A generic 500 is not.
 
-MCP servers that aggregate third-party APIs need careful rate limiting:
+## 3. Rate limiting and cost control
 
-- **Per-agent token budgets** — track and limit usage per connected agent
-- **Upstream API quotas** — queue and prioritize requests when approaching limits
-- **Caching layers** — reduce redundant upstream calls with TTL-based caching
+Production MCP servers get hammered. Agents retry, loops happen, and a single runaway agent can burn your quota. Essentials:
 
-## 3. Observability
+- Per-key rate limits at the capability layer (RPM and daily caps).
+- A free tier generous enough to be useful (we give 1,000 calls/day) but bounded enough to prevent abuse.
+- Cached reads for hot paths. Product search results that don't change second-to-second should come from cache, not from a live merchant scrape.
 
-You can't operate what you can't observe:
+## 4. Distribution: how agents actually find and install you
 
-- **Request tracing** — trace every MCP request through the full stack
-- **LLM-friendly metrics** — expose structured metrics agents can consume
-- **Health endpoints** — implement the MCP health check spec
+A server nobody can install is a server that doesn't exist. The production distribution stack:
+
+- **npm package** for one-command installs (`npx -y @buywhere/mcp-server`).
+- **Official MCP registry** listing for discoverability.
+- **Verified client configs** for Claude Desktop, Cursor, VS Code, Cline, Windsurf, and Codex — the exact JSON each expects.
+- **A self-service key endpoint** so an agent or builder can get credentials without a signup flow.
+
+```json
+{
+  "mcpServers": {
+    "buywhere": {
+      "command": "npx",
+      "args": ["-y", "@buywhere/mcp-server"],
+      "env": { "BUYWHERE_API_KEY": "bw_..." }
+    }
+  }
+}
+```
 
-## 4. Authentication and Authorization
+## 5. Observability: you can't run what you can't see
 
-Production MCP servers need robust auth:
+Track tool call volume, p95 latency, error rates, and per-key usage. When an agent silently stops working, the first question is always "did our server change, or did their call pattern change?" Without per-tool telemetry you're guessing.
 
-- **API key authentication** — validate keys at the transport layer
-- **Capability scoping** — restrict which tools/resources each agent can access
-- **Audit logging** — log every tool invocation for compliance
+## The takeaway
 
-## Conclusion
+The gap between a demo MCP server and a production one is the gap between a script and a service: layered architecture, agent-first tool design, real rate limiting, frictionless distribution, and observability. Get those five right and your server earns daily agent calls instead of sitting in a README.
 
-Building production MCP servers requires thinking beyond the protocol spec. At BuyWhere, we've open-sourced our server patterns so the community can build reliable MCP infrastructure from day one.
+*BuyWhere is a production MCP server searching 288M+ products across SG, SEA, and US markets. Get a key at [buywhere.ai/api-keys](https://buywhere.ai/api-keys) and read the full API at [docs.buywhere.ai](https://docs.buywhere.ai).*
diff --git a/content/blog/buywhere-mcp-goes-live.md b/content/blog/buywhere-mcp-goes-live.md
@@ -1,73 +1,90 @@
 ---
 slug: "buywhere-mcp-goes-live"
-title: "BuyWhere MCP Goes Live"
-description: "BuyWhere's MCP server is now available for all developers. Search 11M+ products, compare prices across merchants, and build shopping agents with the Model Context Protocol."
-author: "BuyWhere Team"
-publishedAt: "2026-07-14"
-tags: ["MCP", "launch", "BuyWhere", "shopping", "AI agents"]
-jsonLd: >
-  {
-    "@context": "https://schema.org",
-    "@graph": [
-      {
-        "@type": "Article",
-        "headline": "BuyWhere MCP Goes Live",
-        "description": "BuyWhere's MCP server is now available for all developers.",
-        "datePublished": "2026-07-14",
-        "author": { "@type": "Organization", "name": "BuyWhere Team", "url": "https://buywhere.ai" },
-        "publisher": {
-          "@type": "Organization",
-          "name": "BuyWhere",
-          "url": "https://buywhere.ai",
-          "logo": { "@type": "ImageObject", "url": "https://buywhere.ai/logo.png" }
-        }
-      }
-    ]
-  }
+title: "BuyWhere MCP Goes Live: The Open-Source Commerce API for AI Agents"
+publishedAt: "2026-07-17"
+excerpt: "BuyWhere is live as a production MCP server — 288M+ products, SG/SEA/US markets, one-command install. Here's what it is, why we open-sourced the client, and how to wire it in."
+tags: ["mcp", "announcement", "open-source", "ecommerce"]
+author: "Lyra"
+jsonLd:
+  "@context": "https://schema.org"
+  "@type": "Article"
+  headline: "BuyWhere MCP Goes Live: The Open-Source Commerce API for AI Agents"
+  datePublished: "2026-07-17"
+  author:
+    "@type": "Organization"
+    name: "BuyWhere"
 ---
 
 # BuyWhere MCP Goes Live
 
-Today we're excited to announce that BuyWhere's MCP server is officially live for all developers. Starting now, any MCP-compatible agent can search, compare, and purchase products across 11M+ items from thousands of merchants — all through a standardized protocol.
+BuyWhere is now live as a production MCP server. One command gives any MCP-compatible agent the ability to search 288M+ products, compare live prices, and surface deals across Singapore, Southeast Asia, and the United States. This post is the short version: what it is, why we built it, and how to wire it in.
+
+## What it is
+
+BuyWhere is a commerce MCP server. It exposes three core tools that return structured, comparable data:
+
+- **`search_products`** — keyword, category, price-range, and market-filtered search across Shopee, Lazada, Amazon, Walmart, Carousell, Qoo10, and more.
+- **`compare_prices`** — side-by-side price and availability for a product across merchants, with a best-value pick.
+- **`discover_deals`** — price-dropped and time-limited offers, filterable by market and category.
+
+Every response is a typed object an agent can reason over and pass onward — not prose to parse.
+
+## Why we built it
+
+Agents got good at reasoning over text and code, but they're blind to live commerce. A model will happily hallucinate a price; what it can't do is tell you whether the Sony WH-1000XM5 is cheaper on Shopee or Lazada *right now*, in SGD, in stock. That data is live, fragmented, and behind anti-bot protections. We built the infrastructure — a deduplicated product index, a merchant scraper fleet, a normalization layer — and exposed it as the standard tool surface agents already know how to call.
 
-## What This Means for Developers
+## Why open-source the client
 
-Building a shopping agent used to mean writing custom integrations for every merchant. With BuyWhere's MCP server, you get:
+The MCP client is open source (`@buywhere/mcp-server` on npm, source on GitHub). Open-sourcing the client means:
 
-- **One integration** — connect once, access thousands of merchants
-- **Standard tools** — `search_products`, `compare_prices`, `get_deals`, and more
-- **Real-time data** — live pricing, availability, and merchant information
-- **Cross-border commerce** — support for US, Singapore, Malaysia, Japan, and more markets
+- **Auditable** — you can read exactly what tools your agent is calling and what data comes back.
+- **Self-hostable** — run it where your agent runs, no mystery relay.
+- **Composable** — fork and extend; the protocol is standard, your tools can build on ours.
 
-## Getting Started in 5 Minutes
+The product index and merchant fleet stay hosted — that's the part that's genuinely expensive to run and where the value compounds — but the interface is open.
+
+## How to wire it in
+
+Three steps, under a minute:
 
 ```bash
-# Install the MCP server via npm
-npm install @buywhere/mcp-server
+# 1. Get a free key (no signup, no email)
+curl -X POST https://api.buywhere.ai/v1/auth/register \
+  -H "Content-Type: application/json" \
+  -d '{"agent_name":"your-agent"}'
+# → {"api_key":"bw_...","tier":"unverified","rate_limit":{"daily":1000}}
+```
 
-# Or use it directly with any MCP client
-# Add this to your MCP configuration:
+```json
+// 2. Add to your MCP client config (Claude Desktop, Cursor, VS Code, Cline, Windsurf, Codex)
 {
   "mcpServers": {
     "buywhere": {
       "command": "npx",
-      "args": ["@buywhere/mcp-server"],
-      "env": {
-        "BUYWHERE_API_KEY": "your-api-key"
-      }
+      "args": ["-y", "@buywhere/mcp-server"],
+      "env": { "BUYWHERE_API_KEY": "bw_..." }
     }
   }
 }
 ```
 
-## What's Next
+```text
+# 3. Ask your agent
+"Find me wireless earbuds under $50 available in Singapore, and compare the top 3."
+→ [search_products] → [compare_prices] → structured recommendations with links
+```
+
+## What works today
+
+- **288M+ products** deduplicated across SG, SEA, and US markets.
+- **Verified clients** for Claude Desktop, Cursor, VS Code Copilot, Cline, Windsurf, OpenCode, Codex, and Continue.dev.
+- **Agent-to-Agent (A2A)** protocol support.
+- **Free tier**: 1,000 calls/day, no credit card.
 
-This launch is just the beginning. We're working on:
+## What's next
 
-- **Purchase completion** — agents that can complete purchases end-to-end
-- **Deal discovery** — proactive deal notification tools
-- **Multi-agent commerce** — coordinating multiple agents for complex shopping workflows
+The commerce layer is live. The next milestones are depth — more merchants, more markets, richer deal detection — and the ecosystem: agents that don't just search and compare but actually complete purchases end-to-end.
 
-## Join Us
+If you're building a shopping agent, a price monitor, or a cross-border comparison tool, the infrastructure is ready. Come build on it.
 
-Try BuyWhere MCP today at [buywhere.ai/mcp](https://buywhere.ai/mcp). We're building the commerce infrastructure layer for AI agents, and we'd love your feedback.
+*Get your free key at [buywhere.ai/api-keys](https://buywhere.ai/api-keys). Full API docs at [docs.buywhere.ai](https://docs.buywhere.ai). Source on [GitHub](https://github.com/BuyWhere/buywhere-mcp).*