How to Improve Your Agent Readiness Score
We scanned 500 businesses. The average score is 43 out of 100. Most businesses are invisible to AI agents or visible but unusable. This is the step-by-step playbook to change that — with real examples, estimated effort for each step, and the exact changes that moved real companies up the leaderboard.
Start Here: Know Your Baseline
The first step is always measurement. You cannot improve what you do not measure, and most businesses have never checked how they appear to AI agents. Our free scan at agenthermes.ai/audit checks your business across all 9 dimensions in about 10 seconds and gives you a score from 0 to 100.
Out of 500 businesses scanned, 250 score Bronze (40-59) and 198 score below 40 — effectively invisible to AI agents. Only 51 score Silver (60-74), 1 scores Gold (Resend at 75), and zero score Platinum (90+). The average is 43. If you have not scanned yet, you are statistically likely to be in the Bronze range or below.
The good news: improving your score is not mysterious. The 10 steps below are ordered by impact-to-effort ratio. The first three steps require no code changes to your product and can be done in an afternoon. Steps 4-7 require moderate engineering work. Steps 8-10 are strategic investments that push you from Silver toward Gold.
The 10 Steps
Run Your Free Scan
Before you improve anything, you need to know where you stand. Go to agenthermes.ai/audit and enter your domain. In 10 seconds, you get a score across all 9 dimensions with a detailed breakdown of what is working and what is not. This is your baseline. Every improvement you make from here is measurable.
Real Example
When we first scanned Stripe, they scored in the low 40s. After their team published better error responses and structured documentation, they recalibrated to 68. You cannot manage what you do not measure.
Publish an OpenAPI Spec
The single highest-impact technical change you can make. An OpenAPI spec tells agents exactly what your API can do, what parameters each endpoint accepts, and what responses to expect. Without it, agents have to guess by parsing documentation pages or, worse, try random endpoints. With it, any agent can auto-generate a working client in seconds.
Real Example
Resend publishes a clean OpenAPI spec and scores 88 on API Quality. Companies without one average 31 on the same dimension. The spec does not need to be perfect — it needs to exist.
Add llms.txt to Your Root
llms.txt is a simple text file at your domain root that tells AI models what your business does, what capabilities you offer, and how to interact with you. Think of it as robots.txt for AI agents — except instead of telling crawlers what to avoid, it tells agents what to use. The format is straightforward: a title, a description, and links to your key resources.
Real Example
A restaurant that adds llms.txt with its menu link, hours, and booking URL goes from completely invisible to discoverable. For SaaS, include your API docs link, pricing page, and authentication guide. This single file is the difference between an agent finding you and skipping you.
Create an agent-card.json
An agent-card.json file at /.well-known/agent-card.json is the A2A (Agent-to-Agent) protocol standard for machine-readable business identity. It contains your business name, capabilities, authentication methods, rate limits, and supported protocols. When an agent encounters your domain, the first thing it checks is this file. If it exists, the agent knows immediately what you can do and how to interact. If it does not, the agent has to scrape and guess.
Real Example
Zero of the bottom 198 businesses in our dataset (the "Not Scored" tier below 40) have an agent-card.json. All 51 Silver-tier businesses have at least partial agent discovery files. Correlation is not causation, but the pattern is clear: businesses that make themselves machine-readable score higher.
Structure Your Pricing
This is the universal weakness across our entire 500-business dataset. D4 Pricing Transparency averages the lowest of all 9 dimensions. The reason: most businesses treat pricing as a human-facing concern — beautiful pricing pages with toggles and sliders that agents cannot parse. The fix is to publish pricing in a structured, machine-readable format. This can be a JSON endpoint, structured data on your pricing page, or a dedicated pricing section in your agent-card.json.
Real Example
Stripe scores 45 on Pricing Transparency despite being one of the best API companies in the world. Their pricing is technically public but requires parsing complex calculator pages. A simple JSON endpoint listing plans, prices, and features would move that score by 20+ points. For smaller businesses: a JSON-LD PriceSpecification on your pricing page takes 30 minutes to implement.
Return Structured Error Responses
When an agent sends a bad request to your API, what does it get back? If the answer is an HTML error page, a generic 500, or an empty response, you are losing points on two dimensions simultaneously. Structured error responses with machine-readable error codes, clear messages, and suggested fixes let agents self-correct without human intervention. Our scanner specifically checks for this: a 401 response with structured JSON scores 87% of what a 200 with valid data scores. An HTML error page scores near zero.
Real Example
Resend returns typed error codes like "validation_error" and "missing_required_field" with specific field names. An agent that hits this error can immediately fix the request and retry. Compare this to a business that returns "Something went wrong" — the agent has no path forward.
Simplify Your Onboarding
D3 Onboarding is the second-weakest dimension across our dataset, right behind Pricing Transparency. The pattern is the same everywhere: CAPTCHA, email verification loops, phone verification, manual approval queues, mandatory credit card entry, multi-step wizards with JavaScript-heavy flows that agents cannot navigate. Every human-required step is a point where agents fail. The goal is not to remove security — it is to provide a programmatic onboarding path alongside the human one.
Real Example
Resend gets you from signup to first API call in under 2 minutes with a generous free tier that requires no payment. That single decision — a free tier with minimal friction — contributes to their Gold score more than any technical optimization. If you cannot offer a free tier, offer a sandbox environment with test credentials.
Add Rate Limit Headers
Rate limiting is expected. But how you communicate limits to agents matters enormously. Returning a 429 status code with Retry-After, X-RateLimit-Remaining, and X-RateLimit-Reset headers lets an agent manage its own request cadence without guessing. This improves both your Security score (you are protecting your service) and your Reliability score (agents can plan around limits). Our scanner checks for proper 429 responses and penalizes APIs that either have no rate limiting or rate-limit silently without headers.
Real Example
Stripe and Resend both return documented rate limit headers. Businesses without them average 12 points lower on D7 Security and 8 points lower on D8 Reliability.
Publish SDKs or Code Examples
Agent Experience (D9) measures how easy it is for an agent to actually use your product after discovering it. SDKs in popular languages, copy-paste code examples, and interactive API playgrounds all contribute. You do not need SDKs in every language — even one well-maintained SDK plus curl examples covers most agent use cases. The key is that an agent should be able to go from "I know what this API does" to "I am making successful calls" without reading prose documentation.
Real Example
Resend scores 82 on Agent Experience with official SDKs in Node.js, Python, Ruby, PHP, Go, and Elixir. But even businesses with a single Python SDK and clear curl examples score 15-20 points higher on D9 than those with only prose documentation.
Activate the Agent-Native Bonus
The Agent-Native Bonus is a 7% weight multiplier that rewards businesses for adopting agent-specific protocols. This includes publishing an MCP (Model Context Protocol) server, supporting A2A protocol, having an AGENTS.md file, or implementing agent-specific authentication flows. This bonus is separate from the 9 core dimensions and stacks on top. Our scanner detects MCP servers, A2A agent cards, AGENTS.md files, and other agent-native signals automatically.
Real Example
Supabase scores 69 partly because of their MCP server, which exposes database operations as agent-callable tools. Any business that publishes an MCP server with even basic read operations gets an immediate scoring boost that competitors without one cannot match.
Priority Order: Maximum Impact in Minimum Time
Not every step is equal. Here is the priority order based on the impact-to-effort ratio, grouped into three phases. Phase 1 can be done in a single afternoon. Phase 2 takes a week of engineering. Phase 3 is a strategic investment over weeks to months.
Phase 1: Afternoon (Steps 1, 3, 4)
Run your scan, add llms.txt, and create agent-card.json. Zero code changes. Expected improvement: +15-25 points.
Phase 2: One Week (Steps 2, 5, 6, 8)
Publish your OpenAPI spec, structure your pricing, fix error responses, and add rate limit headers. Expected improvement: +15-30 points.
Phase 3: Strategic (Steps 7, 9, 10)
Simplify onboarding, publish SDKs, and build agent-native features (MCP server, A2A support). This is what pushes Silver to Gold. Expected improvement: +10-20 points.
Common Mistakes to Avoid
Optimizing your best dimension instead of your worst
Your score is a weighted average. Improving a dimension from 80 to 90 adds less than improving one from 20 to 50. Fix the weakest link first.
Adding agent features without fixing API basics
An MCP server is worthless if your API returns HTML errors. Get the fundamentals right (D2, D6, D7) before chasing agent-native bonus points.
Gating everything behind sales teams
Every sales-gated step is a zero for that dimension. Agents cannot fill out "Contact Sales" forms. Offer a self-serve path, even if enterprise pricing needs human negotiation.
Beautiful pricing pages that are not machine-readable
A pricing page with interactive sliders and tooltips is great for humans and invisible to agents. Add JSON-LD structured data or a /pricing.json endpoint.
Real Results from the Leaderboard
The businesses at the top of our leaderboard are not there by accident. They followed the same principles in this guide — most of them before agent readiness was even a concept. Here is what the top 5 have in common:
Gold. Small API surface, excellent documentation, generous free tier, structured everything.
Silver. Strong real-time API with SDKs in 10+ languages. Loses points on pricing complexity.
Silver. Clean developer experience. Gaps in agent-native discovery (no agent-card.json).
Silver. Public status APIs are inherently agent-friendly. Simple, predictable data.
Silver. Surprisingly strong API docs for a consumer platform. Onboarding is the bottleneck.
The Pattern
Every top-scoring business shares three traits: structured API responses, minimal onboarding friction, and transparent pricing. None of these require cutting-edge AI technology. They require good engineering fundamentals applied consistently across every dimension. The playbook in this guide is the same one the leaders followed — you just have a map now.
Frequently Asked Questions
How quickly can I improve my Agent Readiness Score?
The fastest improvements come from Steps 2-4: publishing an OpenAPI spec, adding llms.txt, and creating agent-card.json. These are static files and specification documents that require no code changes to your core product. Together, they can move your score 15-25 points in a single afternoon. Deeper improvements like onboarding simplification and structured pricing take longer but have outsized impact on your weighted score.
Which dimension should I fix first?
Fix your lowest-scoring dimension first. The math is simple: improving a dimension from 20 to 50 adds more to your weighted average than improving one from 70 to 85. Run your free scan at agenthermes.ai/audit to see your dimension breakdown, then start with the red and orange scores. For most businesses, that will be D3 Onboarding or D4 Pricing Transparency, since these are the weakest dimensions across our entire 500-business dataset.
How is the Agent Readiness Score calculated?
We scan businesses across 9 weighted dimensions: Discoverability (12%), API Quality (15%), Onboarding (8%), Pricing Transparency (5%), Payment (8%), Data Quality (10%), Security (12%), Reliability (13%), and Agent Experience (10%), plus an Agent-Native Bonus (7%). Each dimension is scored 0-100 and the weighted average produces the final score. Tier thresholds: Platinum 90+, Gold 75+, Silver 60+, Bronze 40+, Not Scored below 40.
What moved Stripe from 40 to 68?
Stripe was initially scored in the low 40s in early calibration. The recalibration to 68 came from properly weighting their excellent API quality (D2), strong security posture (D7), and high reliability (D8). Their main gaps remain D4 Pricing Transparency (45) and D3 Onboarding (complex multi-step flow). If Stripe published a machine-readable pricing endpoint and simplified their agent onboarding path, they could reach Gold (75+).
Can I improve my score without changing my product?
Yes. Steps 3, 4, and 5 — adding llms.txt, agent-card.json, and structured pricing data — require zero changes to your actual product or API. They are metadata and discovery files that sit alongside your existing infrastructure. These three steps alone can move most businesses from Bronze to Silver range. Product changes (better error responses, simpler onboarding, SDKs) push you from Silver toward Gold.
Does AgentHermes offer remediation help?
Yes. After your free scan, visit agenthermes.ai/remediate for automated remediation recommendations. We generate the specific files you need (agent-card.json, llms.txt, MCP tool definitions) based on your scan results. For businesses that want hands-on help, our Connect wizard at agenthermes.ai/connect walks you through the full setup step by step.
Ready to start improving?
Run your free scan to see your current score and dimension breakdown. Then use the remediation tool to generate the files you need.