--- title: "My site got 100 in Vercel’s Is Agentic agent-ready scoring tool – here’s how" date: 2026-09-02 modified: 2026-09-07 author: "Miriam Schwab" url: "https://miriamschwab.me/100-score-vercel-agent-ready-heres-how/" markdown_url: "https://miriamschwab.me/100-score-vercel-agent-ready-heres-how.md" type: post excerpt: "Two scanners graded my site on how usable it is by AI agents. The highest-scoring thing I added was a text file saying my site has no login." categories: - "Artificial Intelligence" - "WordPress" tags: - "agent-readiness" - "agentic-web" - "ai" - "mcp" - "plugin" - "wordpress" --- Making sites agent-ready is all the rage, and there are a lot of tools out there for scoring how agent-ready sites are, helping to identify gaps that theoretically should be filled in to make a site a great destination for LLMs. One of those tools is called [Ora](https://ora.ai/), and I saw that Vercel implemented a customized scoring system based on Ora as well called [Is Agentic](https://is-agentic.com/). So although I had already done quite a bit of work to [make my site agent ready](https://miriamschwab.me/how-i-improved-my-sites-ai-readiness-score-from-51-to-86/) according to [Cloudflare’s scoring tool](https://isitagentready.com/), and I had recently determined that AI agents don’t really care about all these things we’re supposed to do to make our sites agent ready and that these scores are mostly vanity metrics ([you can learn more about that in my WordCamp US talk](https://miriamschwab.me/wordpress-ready-for-agentic-web/)), I’m a sucker for getting good grades and decided to tackle this new scoring method as well. So I ran this site, miriamschwab.me, through Ora and got a 56 out of 100 and a grade C. This is actually a decent score compared to most sites, but why settle for C when you can get an A! ![Ora scan result for miriamschwab.me showing 56 out of 100 with a C grade, marked Needs Work and ranked #9952 of 52,442 sites.](https://miriamschwab.me/wp-content/uploads/2026/08/miriamschwab-me-·-56-100-C-Ora-08-27-2026_06_48_AM-copy.png)After iterating based on Ora’s recommendations, I got the score up to 88, which gives me a pretty green A. Note that my site ranks 106 out of 59772 sites analyzed by Ora, which goes to show how few sites are implementing these optimizations. It’s a great time for early adopters to stand out. [![Ora scan result for miriamschwab.me showing 88 out of 100 with an A grade, marked Agent-Ready and ranked #106 of 59,772 sites, with layer scores Discovery 13/20, Access 30/30, Usability 36/40 and Payments 10/10.](https://miriamschwab.me/wp-content/uploads/2026/08/Monosnap-miriamschwab.me-·-88100-A-Ora-2026-08-31-14-18-41.png)](https://ora.ai/score/miriamschwab.me)Vercel launched [Is Agentic](https://is-agentic.com/), which is a modified version of the Ora scanner based on the parameters they think are important or make sense for most sites. Overall, their scores are based on more logical, general features that do make sense to implement and can possibly even add value. If you want a useful analysis where the gaps are reasonable to implement, check Is Agentic out. My site scores 100 there, which of course gives it brownie points in my eyes 🤩 [![Is Agentic scan result for miriamschwab.me showing 100 out of 100 and the verdict “Strong technical baseline”, with Essential 78.6/80, Recommended 16.7/20 and 5 bonus points.](https://miriamschwab.me/wp-content/uploads/2026/08/image-2-1024x373.png)](https://is-agentic.com/scan/miriamschwab.me)To achieve these high scores, I implemented a lot of the recommendations, some of which add questionable value – for example, the highest-scoring single thing I added for Ora was a text file that says my site has no login. What a time to be alive! But I have to admit that some of the features implemented actually turned out to be useful, which was surprising. ## What the scanners actually check Both tools explore a site the way an agent theoretically would (although as I mentioned, my testing showed that agents mostly explore sites like humans do, at least for now). They look for an OpenAPI spec, an MCP server, machine-readable errors, a markdown version of each page, structured data, and a dozen conventions with names like ARD and NLWeb that most people building websites have never heard of. Ora scores across four layers: discovery, access, usability, and payments. Vercel’s model runs on Ora’s audit evidence, so while the scoring differs slightly, improvements move the scores in both up together. My site already did some of this agent-ready stuff. It’s been serving llms.txt, llms-full.txt, and a `.md` twin of every page for months, through my [Make My Site Agent-Ready plugin](https://miriamschwab.me/plugins/make-my-site-agent-ready/). However, while my plugin addressed the question “can an agent read you”, it didn’t address all the other things related to “can an agent call you.” ## The audit identified some new agent-ready features My site was missing a few “key” things for getting a high score with these tools: one was an OpenAPI spec, so nothing told an HTTP client how to construct a request, and the other was an MCP server, so an agent that speaks the [Model Context Protocol](https://miriamschwab.me/why-mcps-are-the-hooks-of-the-ai-era/) had to fetch and parse files (which my testing showed it does really well, thank you very much). Both now exist, mostly to please the scoring overlords. The OpenAPI document at `/openapi.json` (documented on [my API page](https://miriamschwab.me/api/)) is automatically generated from what the site actually serves: it reads the feature toggles, asks WordPress which REST routes are really registered on this install, and describes only those. The MCP server is public, read-only, and needs no authentication. It exposes four tools: search the content, list it, read any page as markdown, and get an overview of the site. It reaches nothing that llms-full.txt doesn’t already publish, and it’s rate limited to 60 calls a minute per IP, because unlike a static file it answers by running a database query which can get heavy if abused. ## The file that scored highest Ora has a cluster of checks about authentication: OAuth 2.0 support, scoped permissions, protected-resource metadata, an auth.md document, agent auth discovery. Together they’re worth more than 20 points. That’s a lot of tasty points! However, my site has no authorization server, no API keys, and no accounts. The tempting move is to publish the metadata documents anyway and improve my score, but I decided not to. I could theoretically pass those checks dishonestly by either creating fake endpoints or publishing links to 404s, and while I implemented a lot of things of questionable value to get my score up, lying about this just to get points was too much. These factors should not be required for their scoring method since they’re not relevant to all sites. Requiring them to get a good score just encourages site owners to fake it on their site, which isn’t good for humans or machines. What I published instead was `/auth.md`, a plain walkthrough of how an agent gets access to my site. This is an example of a questionable feature because the answer is that it doesn’t need to 😂. Everything on my site that is readable is open, sends no Authorization header, and you’ll never get a 401 because there’s nothing here to be unauthorized for. The one endpoint that does take an action, [my contact form](https://miriamschwab.me/my-contact-form-is-now-an-agent-ready-api/), uses a single-use token you fetch from the endpoint itself right before you post, and the file explains that flow in the order you’d do it. But at least implementing it didn’t involve adding misleading information, so I figured I might as well. That shallow document added about 10 points to my score 🎉. It passed the two auth.md checks outright, and it flipped OAuth 2.0 support and scoped permissions from partial credit to full. Clearly describing that I have no authentication scored very nicely. ## The scanner did find some real issues, so it was actually somewhat useful Ora pointed out a real inconsistency: two of my custom post types, Press & Talks and Plugins, had no markdown versions. [My plugin](https://miriamschwab.me/plugins/make-my-site-agent-ready/) was configured for posts and pages only, so 45 pieces of published content were invisible to every agent-facing file on the site. They were in the sitemap and in search results, but absent from llms.txt, llms-full.txt, and the MCP server. I fixed that with one checkbox in my plugin settings, and llms.txt grew from 12KB to 27KB. Ora also noticed that the posts on my site had inconsistent authors. My markdown files carry YAML frontmatter with an author field, and in 53 items my agent WordPress user was listed instead of me! So I updated the authors across my posts to be consistent, which did make my site better overall. ## The points I sacrificed for the sake of honesty Counting the auth cluster, I left about 28 points’ worth of checks unclaimed, because they were just too irrelevant to my site that adding them would be beyond ridiculous: a Web Bot Auth directory which needs real Ed25519 keys and signature verification; a sandbox environment which needs a sandbox; multi-language SDK packages which need SDKs. Then there’s the demand for a Wikipedia presence which needs third-party press coverage first. That’s not something you can just implement. ## Despite my cynicism, these changes do make my site better in some ways I must begrudgingly admit that some of these changes didn’t just give me a higher shiny score, they actually improved the quality of my site. Adding an OpenAPI spec makes my site callable, a 404 that returns a JSON body pointing at the sitemap gives an agent somewhere to go instead of a dead end, and author metadata with my name on it that is correct is definitely preferred. If you too want to get higher scores, know that all of the above ships in the latest version of my [Make My Site Agent-Ready](https://miriamschwab.me/plugins/make-my-site-agent-ready/) plugin. If you run it against your own site, I’d like to hear what your score does!