Development · Framework
Agent-Ready SaaS Checklist
Practical rules for preparing SaaS APIs and infrastructure for AI agents
An opinionated architectural checklist by Rhys Sullivan outlining how SaaS products should expose their functionality to autonomous AI agents. It advises separating raw API endpoints from Model Context Protocol (MCP) workflows, writing strict OpenAPI specs, and avoiding unnecessary CLIs in favor of solid auth.
The framework
- APIs and MCPs are different and you need both.
- APIs: give agents raw access to actions and data.
- MCPs: for flows (e.g. adding an environment variable to a Vercel project, which needs a browser to input securely).
- Build on OpenAPI; clear descriptions, examples, documented errors.
- Convert human docs to 'skills' (docs for agents) by swapping 'Click this button in the dashboard' for 'call this API'.
- Enable CIMD OAuth for both MCP server and API server.
- No CLI unless your service touches local files — agents call the API/MCP directly; solve auth instead.
- 90% of this already exists in most apps; focus on primitives: OpenAPI, OAuth.
What it's for
- Architect SaaS integrations specifically targeted at autonomous AI agents
- Decide when to build an MCP server versus exposing a standard REST API
- Adapt existing developer documentation into machine-readable agent skills
- Prioritize authentication workflows like CIMD OAuth for non-human clients
Key points
- Differentiates APIs (raw data and actions) from MCPs (complex multi-step flows)
- Emphasizes clean OpenAPI specs with explicit descriptions, examples, and documented errors
- Provides a strategy to transform human UI instructions into direct API-call 'skills'
- Recommends CIMD OAuth support across both API and MCP servers
- Advises against building dedicated CLIs unless local file access is strictly required
- Focuses engineering effort on reusable primitives rather than custom agent wrappers
Pricing
Free
Details
- Type
- Framework
- Category
- Development
- Topic
- AI Development
- Open source
- No
- Platforms
- API, other
- For
- SaaS founders, software architects, and backend engineers building for AI agents
- Tags
- ai agents, mcp, api design, openapi, oauth, saas
Why it was saved
- @RhysSullivan: Concrete, opinionated steps on what to build (API + MCP, OpenAPI, OAuth/CIMD) and what to skip (CLIs) when preparing software for agent use.
Related
Turn websites and documents into clean, LLM-ready markdown or structured data.
Turn any website into a structured REST API using natural language
Agentic canvas and IDE extension for building UI designs directly into code
Connects AI coding agents to browser annotations for automated UI reviews
Agent-first control plane backed by 2,400+ curated AI developer skills
Turn websites into plain HTTP APIs for AI agents to call directly.
Updated