Industry deep dive

Agent readiness for Software / SaaS / Developer Tools

How AI agents discover, understand, and recommend saas businesses — and the specific signals we check when scanning a saas site.

5 min read· Updated 2026-04-25

What agent-ready means for SaaS websites

Agent-ready means an AI can discover your product's capabilities, read your pricing, consume your API docs, and—if you ship an MCP server—install and call your tool without human mediation. For SaaS, this translates directly to measurable outcomes: an agent evaluating six project-management platforms can parse your /standards/llms_txt quickstart, compare your API limits in a machine-readable OpenAPI spec, and surface your integration examples in a side-by-side grid for a VP of Engineering.

It's not theoretical. Claude Desktop agents already spin up local SDKs using MCP servers. ChatGPT's research mode scrapes /api/openapi.json endpoints to autogenerate code samples. Cursor agents pull Markdown-negotiated docs to inline setup instructions in the IDE. If your SaaS lacks these artifacts, you're invisible in the decision loop—no agent citation, no RFP shortlist inclusion, no "install this tool" suggestion in a CTO's chat thread.

Why AI agents matter for SaaS businesses in 2026

ChatGPT introduced Browse with Bing in 2023, but Q4 2025 brought the inflection: ChatGPT Shopping, Claude's MCP ecosystem, and Perplexity's Pro Search all began returning actionable tool recommendations in conversational responses. Stripe's OpenAPI spec is indexed by every major agent runtime. Linear's public MCP server lets agents create issues programmatically during planning sessions. Vercel ships an llms.txt that routes agents straight to framework-specific deployment docs. These aren't experiments—they're distribution channels.

The business outcome is citation rate. When a procurement agent compares analytics platforms, PostHog appears in the final three because its docs pages serve Markdown on Accept: text/markdown requests and its /llms.txt maps quickstart → integration → troubleshooting in 12 lines. Competitors with PDF whitepapers and JS-gated docs don't surface. Agent-driven discovery isn't replacing Google; it's replacing how engineers and product teams research tools. If your docs aren't parseable, you're out.

The 4 standards that move the needle for SaaS

MCP server card — A machine-readable manifest (JSON or YAML) declaring your MCP server's install command, schema version, and tool endpoints. Agents use this to add your product to their context without OAuth handshakes or manual setup. Stripe's stripe-mcp-server npm package is the reference implementation.
OpenAPI spec at /api/openapi.json — A complete OpenAPI 3.1 document describing every endpoint, parameter, and response schema. Agents parse this for tool discovery, autogenerate SDKs, and compare rate limits. Host it at a stable, public URL; link it in your docs footer and <head> with <link rel="service-desc">.
Markdown content negotiation on docs — Serve raw Markdown when agents send Accept: text/markdown. No JS frameworks, no React hydration—just clean .md source with frontmatter. This lets agents extract code blocks, parameter tables, and headings without running Puppeteer.
llms.txt mapping — A root-level /llms.txt file that routes agents to your quickstart, API reference, and integration guides in priority order. One URL per line, prefixed with # Quickstart, # API Reference, etc. Vercel's version is 18 lines and covers 90% of agentic onboarding queries.

Common gaps we see on SaaS sites

PDF-only API specs — A 47-page PDF titled "API_Reference_v3_Final.pdf" is unreadable to agents. Publish OpenAPI JSON.
JS-gated pricing tables — React components that render pricing tiers client-side return blank HTML to agents. Use server-rendered <table> with itemprop="offers" schema.
Missing dateModified on changelog — Agents check doc freshness via <meta property="article:modified_time">. Stale timestamps = stale trust.
No MCP install path — You ship an SDK but no MCP server card. Agents can't autodiscover or install your tool in chat contexts.
Broken /api/openapi.json redirects — The spec exists but lives at /docs/api/v2/spec.yaml with no canonical redirect. Agents check /api/openapi.json by convention.

How to test your SaaS site for agent readiness

Run an HTTP request against your docs with Accept: text/markdown and see if you get usable content or a React error boundary. Check if /llms.txt exists and points to real pages. Verify that /api/openapi.json returns a valid spec and includes rate-limit headers. Test whether your homepage includes schema.org/SoftwareApplication structured data with applicationCategory and offers.

Or skip the cURL one-liners and run a free scan — we'll grade your site across 25+ deterministic checks weighted for SaaS, from MCP discoverability to FAQ schema to pricing-table markup. You'll get a percentile score against 1,200+ developer-tool sites and a prioritized fix list.

FAQ

Do I need an MCP server if I already have a REST API?

Yes. OpenAPI helps agents discover your endpoints; an MCP server lets them call those endpoints in conversational context without leaving the agent runtime. Think of MCP as the npm-install layer—agents can add your tool to their workspace in one step. Stripe, Linear, and Supabase all ship both.

Will exposing OpenAPI specs increase scraping or abuse?

Public specs don't bypass rate limits or auth. Agents still hit your API with the same credentials and quotas as browser clients. The spec is metadata—it tells agents what exists, not how to bypass your firewall. Stripe's spec has been public since 2019 with no uptick in abuse.

Can agents actually parse our Notion-based docs?

No. Notion pages render as JS blobs; agents scraping them get loader spinners. Migrate your quickstart and API reference to Markdown files served via content negotiation, or publish parallel .md versions in a /docs repo. Linear moved off Notion in Q3 2024 for exactly this reason.

How do we compare to other SaaS tools in agent readiness?

Run a scan and check your percentile. Top-quartile SaaS sites (Stripe, Vercel, PostHog) score 80%+ because they ship all four signals: MCP card, OpenAPI, Markdown negotiation, and llms.txt. Median SaaS sits at 42%—usually missing MCP and content negotiation but present on OpenAPI.

Which SaaS companies already rank high for agent-readiness?

Stripe leads with a public MCP server, OpenAPI 3.1 spec, and Markdown-negotiated docs. Vercel ships llms.txt and schema-complete pricing tables. Linear publishes an MCP install command in their footer. PostHog serves raw .md on Accept: text/markdown. Supabase maintains a 2,100-line OpenAPI spec at /api/openapi.json. All five appear in the top 8% of scanned SaaS sites.

How long does it take to implement the core four standards?

MCP server card: 1 sprint if you have an SDK; 3 weeks if you're building from scratch. OpenAPI spec: 2-4 weeks depending on endpoint count and existing docs. Markdown negotiation: 1-2 days if your docs are already in .md; 1-2 weeks if you need to reverse-engineer them from a CMS. llms.txt: 30 minutes. Front-load MCP and OpenAPI; they unlock the most agentic distribution.

Test it on your saas site

We grade saas sites against 25+ deterministic checks.

One scan. Per-finding evidence. Free.

Run a free scan