Topic
Calling MCP servers from your own code, connecting Claude to Amazon Ads and Seller Central, and using Claude on listing work.
4 guides, written from real builds.
MCP is sold as the way to plug tools into an AI assistant, so most people assume they need a model in the loop to use one. They don't. An MCP server is a JSON-RPC service with a discovery method, and for some Amazon-adjacent data it is now the only programmatic way in: Helium 10 has no public REST API, and Amazon runs MCP servers for both the Ads API and the Selling Partner API.
This hub collects what I've built and broken on that layer. One thing to know as of October 2026: the MCP specification's current revision, 2026-07-28, drops the initialize handshake that servers built to earlier revisions still expect, and each post says which revision its code targets.
When the job is bigger than a prototype — read-only guardrails, an audit trail, accounts other people depend on — that is the work behind Ads API consulting.
Claude & MCP
If you have gone looking for a Helium 10 REST API to pull Cerebro or Magnet data into your own tool, you already know it does not exist. What does exist, since July 2026, is an MCP server — and you can call it straight from your own code, without Claude or ChatGPT anywhere in the loop. Here is the whole thing: discovery, OAuth, the tools, and the two gotchas that cost the most time.
Claude & MCP
MCP is marketed as the way to give AI assistants access to your tools, so most people assume you need a model to use one. You don't. An MCP server is a JSON-RPC service with a discovery endpoint, and calling it directly is faster, deterministic and free. Here's when to skip the model entirely, when not to, and the whole client in about forty lines.
Claude & MCP
Most AI listing tools start from a title and a few keywords, which means they describe a product nobody has looked at. Sending the actual product photographs to a vision model changes the output — it can see the material, the finish, the scale and what's in the shot. Here's the pipeline, the compliance gate that has to sit after it, and the parts that still need a human.
Claude & MCP
Amazon runs two MCP servers — one for Ads, one for the Selling Partner API — that let Claude query your accounts directly. The built-in connector flow fails with '400 Bad Request: An unknown scope was requested'. Here are the setup steps that work, then why the connector is broken and what the connection costs you in context.
Shahzeb Khan runs Databaaba, a one-person studio that builds claude, mcp and amazon data integrations for sellers, agencies, and software teams. Every guide here came out of a real client build.