Overview

This reference covers the Putty Engine's HTTP API and its MCP surface, in three kinds of page that sit side by side: written pages, the MCP tools grouped by category, and one generated page per HTTP operation.

What this reference coversCopy Anchor Link

The Engine is reached over HTTP. Every route accepts a request that the server validates against a schema and returns either a result or one of the failures the route documents. Each route is described that way here — the fields it accepts, the status codes it can answer with, and the shape of the body it sends back — so a client can be written from these pages without reading the server.

The same engine also speaks the Model Context Protocol, which is how an AI agent reaches the same capabilities. That surface has 82 tools across 11 categories, and the MCP section below covers it one category at a time.

Nothing on these pages is invented. The generated pages are rendered from the OpenAPI document the API app publishes — currently version 0.17.0, 121 paths, 159 operations, 28 tags — so they cannot drift from the routes the server registers. Regenerate it with pnpm --filter api openapi:generate; the docs build reads the committed copy, so a route that is not in the snapshot has no page here.

How the sidebar is organisedCopy Anchor Link

• Getting started — run the whole stack locally, from a fresh checkout to a working API.
• Quick start — the deployed API: mint an sk- key and make your first call.
• The free layer — the thirteen operations that need no bearer at all.
• The seven Kits — what each Kit does today, and where its surface lives.
• Inference — the OpenAI-compatible /v1 surface.
• MCP — connecting an agent, then the tools grouped by category.
• API Reference — the generated pages, one per operation, grouped by the tag the operation carries on the server. API Reference is a folder, not a page, so it has no index URL of its own: the link above points at the first page under Account, and every tag folder sits under API Reference in the sidebar.

Every page has a Markdown copy behind the buttons above its body, and any page can be fetched as Markdown by suffixing its URL with .md.