Free tool
Your API as a connector for Claude and ChatGPT
Your team already buys Claude or ChatGPT. The next question is how to let it work with your own system: orders, customers, invoices, the ERP. Paste your OpenAPI spec and get a proposal of narrow MCP tools, with the irreversible ones flagged, and a server skeleton to start from.
How tools are proposed
Each operation becomes one tool with a name allowed by the MCP specification (letters, digits, underscore, hyphen and dot). GET operations are marked read-only. DELETE and operations about payments, refunds, cancellations or transfers are marked irreversible and left unselected. The input schema comes from path and query parameters and the JSON request body.
What the skeleton is
A starting point for the official TypeScript SDK: one registerTool call per selected tool, a zod input schema and a fetch to your API with a token from an environment variable. Before people use it, it needs your own login per user, limits, logging and tests. That is the part I build.
What it does not do
It does not call your API, does not store your spec and does not judge your API. Everything happens in this tab.
Questions
What is an MCP connector?
An MCP server that exposes tools to AI assistants such as Claude and ChatGPT. Your team adds it once, and the assistant can then look up orders or create a draft invoice in your system, asking before anything irreversible.
Why not expose every endpoint?
Assistants choose better from a short list of well-described tools. Start with the few questions your team asks every day, then add more.
Does my spec leave the browser?
No. Parsing and code generation run in this page. Analytics records only how many operations of each kind the spec had.
Can you build and run it for us?
Yes: the connector on your existing login, deployed for your team, with usage measured per person. The configurator shows the price and the start date.
Sources
Want your system inside Claude and ChatGPT?
I build the connector on your existing login, with narrow tools that ask before anything irreversible, deploy it for your team and measure who uses it. Then new tools as your processes grow.
