Developer API & integrations · AI agents & automation
Machine-readable specs: OpenAPI, agent manifest, and tool catalog
Three public, always-current endpoints describe the entire API for machines: an OpenAPI 3.1 spec, a rich agent manifest with safety flags, and a plain tool catalog. Discovery is open; execution still needs your key.
Why this exists
Modern automation tools and AI agents don't want PDF documentation — they want a machine-readable description they can load and act on. Your platform publishes three, all generated live from the same tool registry that powers the API, so they can never drift out of date.
The three endpoints
OpenAPI 3.1 spec at https://your-domain.com/openapi.json (also served at /.well-known/openapi.json) — the industry-standard format that API clients, code generators, and no-code tools understand.
Agent manifest at https://your-domain.com/.well-known/autocreator-tools.json — the richest view, built for AI agents. Every tool comes with its argument and result schemas, the scope it requires, a read/write/generative classification, destructive and agent-safe flags, whether human confirmation is recommended, its rate bucket, a suggested command-line name, and a ready-to-copy request example. It also includes safe-agent defaults (start read-only, confirm destructive actions, never log keys) and the full scope list.
Tool catalog at https://your-domain.com/ai/tools — a simple JSON list of every API-exposed tool with descriptions and parameters, handy for a quick scan.
Open discovery, protected execution
All three endpoints are public and need no key — like most major API providers, we treat documentation as marketing, not a secret. Actually calling a tool always requires a per-platform API key with the right scope, so open discovery gives away nothing about your data.
Point a CLI generator, an MCP server builder, or an AI agent at the manifest and it can teach itself your entire API in one request. See the guide on connecting your own AI agent for a walkthrough.
Good to know
- The manifest's destructive and agent_safe flags are there so agents can ask before doing anything irreversible — build your automations to respect them.
- These endpoints work on your own custom domain, so agents you connect never need to know the platform behind your brand.
Related articles
Connect your own AI agent to your platform
Prefer your own external AI agent over the built-in assistant? Point an open-source generator like CLI Printing Press at the public spec to produce a command-line tool, agent skill, or MCP server — powered by a scoped key you control.
Help AI assistants discover your site with llms.txt
Your site automatically publishes llms.txt and llms-full.txt — a concise, AI-friendly map of your public pages, programs, collections, and blog articles — so AI assistants can find and reference your content. No setup needed.