Best Agent Readiness Check Tools in 2026
Jun Liang Lee12 min readDirect answer
An agent readiness check has two layers, and Lightsage covers both: its Agent Readiness Audit crawls every page of your site and docs, and its evals run 12 AI coding agents on your task. For a free static scan, use Cloudflare's Is Your Site Agent-Ready, and for a free public baseline, check your tool on Devtool Arena.
What is an agent readiness check?
An agent readiness check is a test of whether an AI agent can find, understand and use your developer product without a person helping. It has two layers. The static layer asks whether your site and docs expose what AI agents need, such as crawler access, content that shows without JavaScript, llms.txt and an API spec. The task layer asks whether a real AI coding agent can finish your first task.
The term gets confused with two others. It isn't a company's AI-adoption readiness, the kind of maturity model AWS and Microsoft publish to help a business decide whether it can deploy AI. It also has nothing to do with human support agents. Here the "agent" is software acting for a developer, such as Claude Code, Codex or Cursor.
You need both layers because a clean static scan doesn't predict a finished task. On Devtool Arena, Lightsage's public leaderboard, 46 of 155 tools on the Claude Code API board had a perfect Agent-Readiness Checklist score of 100. They averaged 71.7 on the task, against 62.5 for the 31 tools scoring under 70, and still ranged from 39 to 91 (our analysis, October 2026 batch).
How we ranked agent readiness check tools
The axis is how much of both layers a tool covers, measured on real pages and real AI agents rather than on a self-reported form. We read each vendor's own pages on November 1, 2026.
- A tool that checks pages and also runs AI agents ranks above one that does only one of the two.
- A tool that checks every page of a site ranks above one that scores a single URL.
- A tool that runs more named AI coding agents ranks above one that runs fewer.
- Where a page is silent, we write "doesn't document". That isn't the same as "doesn't have".
The task layer matters because the static layer explains little of the outcome. Across the 143 Claude Code API board tools with a public company page, 111 (78%) had an OpenAPI spec, and they averaged 69.8 on the task against 70.0 without one. A typed SDK (70.5 against 63.6, n=15) and an MCP server (70.5 against 64.5, n=17) showed bigger gaps, and llms.txt was near-universal at 140 of 143 (98%). On the Gemini, Hermes and Kilo Code boards, full-checklist tools scored no higher than the rest (Devtool Arena API boards, October 2026).
Best agent readiness check tools at a glance
| Tool | Best for | Static layer | Task layer | Where it falls short | Starting price (checked November 1, 2026) |
|---|---|---|---|---|---|
| Lightsage | Both layers across a whole site and docs | Agent Readiness Audit crawls every page, up to 2,500 on Scale | API, CLI, MCP, Custom and Skills Evals on 12 AI coding agents | Readiness and agent runs need Scale; suggests fixes but doesn't host docs | Free tracks 10 ChatGPT prompts a week; Scale from $100/month |
| Devtool Arena | Seeing how your tool compares in public | 7-asset Agent-Readiness Checklist per tool | Public task results, tool calls, runtime and cost | Lightsage's own leaderboard; you get a graded listing, not a private audit | Free |
| Postman AI-ready APIs | An API team that lives in Postman | Scans OpenAPI spec, docs, collection, MCP server, CLI and SDKs | Says it shows how Claude, ChatGPT and Cursor can build with your APIs | Doesn't describe how the agent tests run | Doesn't document a price |
| Built for Devs | A credit-priced static check, with an agent evaluation behind it | Credit-priced static readiness check | Full evaluation runs an agent as an ICP developer | Waitlist; doesn't document which agents run | Pricing shows TBD |
| agent-ready.dev | Weekly monitoring of site readiness | 40 spec checks, 10 llms.txt checks, 15 protocol checks | None documented | No agent ever runs | Free; Pro $19/month |
| Cloudflare Is Your Site Agent-Ready | A fast, free first scan of one URL | Checks across 5 categories | None | One URL at a time; no agent runs | Free, no signup |
| Lighthouse agentic browsing | A CI gate on a single page | Deterministic audits of WebMCP, accessibility and llms.txt | None | Pass ratio, not a score; one page per run | Free, open source |
| Microsoft Scope | A free task test on GitHub Copilot and Claude Code | None | Same task and criteria across agents, with repeated runs | Two agents supported; no site scan; you write the tasks | Free, MIT licence |
1. Lightsage: audits every page, then tests with 12 AI coding agents
Lightsage measures how coding agents and answer engines discover, choose and use developer tools, and it covers both readiness layers in one product. Its Agent Readiness Audit crawls every page of your website and your docs, in separate Website and Documentation modes. Per page it checks AI crawler access in robots.txt, whether content shows without JavaScript and matches the JavaScript version, whether the page returns Markdown when an AI agent asks for it, page length, load time, and the title, description, canonical, JSON-LD and sitemap. It also checks that pricing, docs, API reference and quickstart can be reached.
It ranks fixes by severity, gives a copyable prompt for your coding agent per fix, and can draft an improved page. A Benchmarks view runs Cloudflare's scan inside Lightsage and saves the results. For the task layer, Agent Usability runs real coding agents on your task against your API, CLI or MCP server in sandboxes: pass or fail, traces, errors, and per-agent and per-endpoint scores. Skills Evals runs the same task with and without your agent skill and compares the two, which is how you check a SKILL.md. The 12 agents are Claude Code, Codex, Cursor, GitHub Copilot, Antigravity, OpenClaw, OpenCode, Hermes, Pi, Kilo Code, Grok Code and Muse Code.
The limitation is access. Agent Readiness and coding-agent runs need a paid Scale plan, and Free only tracks 10 ChatGPT prompts a week. Lightsage measures and suggests fixes. It doesn't host your docs or publish the changes for you, so someone on your team ships them.
Price (checked November 1, 2026): Free $0, Scale from $100/month with every coding agent and answer engine in every tier, Enterprise custom (pricing). A 7-day trial includes 1,000 credits and every agent with no card. Coding-agent runs use usage-based credits. The agent experience page covers the wider product.
Best for: devtool teams that want one tool to check every page and then watch real AI coding agents try to use the product.
2. Devtool Arena: a public checklist and task results
Devtool Arena is Lightsage's free public leaderboard at devtoolarena.com, so weigh this entry with that in mind. It benchmarks how coding agents use 150+ developer tools. Each company page shows an Agent-Readiness Checklist of 7 assets (llms.txt, Context7, an MCP server, a typed SDK, an OpenAPI spec, agent skills and a CLI), plus the task prompt, tool calls, runtime and cost. There are API boards for 12 coding agents, and MCP and CLI boards for Claude Code and Codex.
It's the quickest way to see both layers for your category and your rivals before you buy anything. The homepage invites you to "Get your API graded", and you can submit a tool at devtoolarena.com/add. Its Agent Landscape ranks tools by how often coding agents recommend them.
The limitation is that it's a public listing. You get a graded task result on a fixed scenario, not a crawl of your own docs or a private run on your task.
Price (checked November 1, 2026): free.
Best for: founders and DevRel leads who want to see where their product and their rivals stand before they commit to a tool.
3. Postman AI-ready APIs: a scan for API teams
Postman's AI-ready APIs page asks whether your APIs are ready for AI agents. It says it scans your public API ecosystem and returns a graded, evidence-backed report in minutes. The surfaces it lists are your OpenAPI spec, docs, Postman collection, MCP server, CLI and SDKs, and it names machine-readable errors, typed schemas and discoverable auth as things it looks at.
The page says it shows how well Claude, ChatGPT and Cursor can find, read and build with your APIs. It doesn't describe how those tests run, which agent versions are used, or how many runs back a grade.
Price (checked November 1, 2026): the page doesn't document a price. You can configure your organization, book a demo or get your grade.
Best for: API teams already working in Postman that want a graded report on a spec, docs and SDKs.
4. Built for Devs: a credit-priced static check with an evaluation behind it
Built for Devs sells developer growth tools. Its AI Agent Product Evaluations tool has two parts. A credit-priced readiness check reviews llms.txt availability, documentation access, MCP presence, setup instructions and machine-readable product context. The page says it "does not launch a browser session or complete a product task". A paid full evaluation then runs an agent, configured to act as a developer in your ideal customer profile, through discovery, signup, setup and a real build task, and shows the commands, errors and outcome.
That is the two-layer idea in a single product. The limits are practical. Both pages point to a waitlist, the pricing page shows TBD for each plan, and the pages don't document which coding agents run the evaluation.
Price (checked November 1, 2026): pricing shows TBD per month. Top-ups are priced at one credit per dollar.
Best for: dev tool companies that also want positioning and messaging work around the agent findings, and can wait for access.
5. agent-ready.dev: weekly monitoring of site readiness
agent-ready.dev scores a site against the Vercel Agent Readability Spec (15 site-wide and 25 per-page checks), 10 checks on the llms.txt format and 15 on agent protocols such as MCP, A2A and Web Bot Auth. Generators produce files such as llms.txt for you. Its site also lists an MCP server scanner and validators for Agent Cards.
It's a static tool. No AI agent attempts a task, and the pages don't document one. The Free plan scans 25 pages at a time and Pro scans 250, so a large docs site gets sampled.
Price (checked November 1, 2026): Free $0 (25 pages per scan, 10 scans per 30 days). Pro $19/month adds 250 pages per scan, weekly monitoring, email alerts and a REST API (pricing, last updated July 18, 2026).
Best for: a small team that wants a cheap, scheduled alert when a deploy breaks its llms.txt or robots rules.
6. Cloudflare Is Your Site Agent-Ready: the fastest first scan
Is Your Site Agent-Ready? scans a URL against emerging standards in five categories: Discoverability, Content Accessibility, Bot Access Control, Protocol Discovery (MCP Server Card, Agent Skills, OAuth discovery) and Commerce (x402, ACP). Its output includes instructions you can paste into your coding agent to improve the score. The site exposes a scan API at /api/scan.
It checks standards, and it checks the URL you give it. No AI agent attempts a task, and the page doesn't document a whole-site crawl.
Price (checked November 1, 2026): free to use with no signup; the page lists no paid plan.
Best for: anyone who wants a free first look at one URL in a minute.
7. Lighthouse agentic browsing: a CI gate for one page
Lighthouse has an Agentic Browsing category of deterministic audits for machine interaction. The audits cover registered WebMCP tools, forms missing declarative WebMCP, WebMCP schema validity, accessibility for agents, layout stability and an llms.txt audit. The audits are reproducible, so they suit CI.
It has no 0 to 100 score. The category shows a pass ratio, because the standards are still emerging, and a run covers one page. Nothing runs an AI agent against your product.
Price (checked November 1, 2026): free. Lighthouse is open source and ships in Chrome DevTools.
Best for: front-end teams that want a repeatable accessibility and WebMCP check on key pages in CI.
8. Microsoft Scope: a free task test on two agents
Microsoft Scope is an open-source "agentic experience evaluation platform" in research preview. You give coding agents the same task, define what good looks like, and compare the evidence over repeated runs. Criteria are reusable and run in ordered gates, so a failed build skips the tests that depend on it. It targets product surfaces, skills, MCP servers and extensions.
The site lists GitHub Copilot and Claude Code as supported today, OpenAI Codex and OpenCode as planned, and Cursor as not integrated. It doesn't scan your site, so it has no static layer, and you write the tasks and the criteria yourself. The GitHub repository shows an MIT licence.
Price (checked November 1, 2026): free and open source. The pages don't document a hosted service.
Best for: teams that want a free, repeatable task test on GitHub Copilot and Claude Code and are happy to write their own criteria.
Which should you choose?
Match the tool to who is asking and which layer is missing.
| You are | Choose |
|---|---|
| A founder with no budget yet | Cloudflare's scan or agent-ready.dev for the static layer. Then give a coding agent your first task by hand, and look at your Devtool Arena page |
| A DevRel team | Lightsage, for every page of your docs and 12 AI coding agents on your task |
| A developer experience team | Lightsage for a ranked fix list per page and per agent; agent-ready.dev Pro or Lighthouse to catch regressions in CI |
| An API-first CTO | Lightsage API Evals for task results, Postman's scan for a graded read of your spec and SDKs |
| A team shipping an MCP server | Lightsage MCP Evals; the MCP server testing tools list covers more |
Skip Lightsage if you only need a one-off look at a single URL, or if no coding agent touches your product yet. As an illustration, an auth company like Clerk might run the Agent Readiness Audit on its docs, then run API Evals to see which agents stall at sign-in.
For more, see best agent experience tools, best agent testing tools, how to run agent evals and, for a command-line product, how to make your CLI agent-friendly. The free API Agent-Readiness Audit playbook walks through the manual version.
Frequently asked questions
Our DevRel team wants to run an agent readiness check. What tool should we use?
Start with a free static scan, such as Cloudflare's Is Your Site Agent-Ready or agent-ready.dev, then run a task test with real AI coding agents. Lightsage crawls every page of your site and docs and runs 12 AI coding agents on your task, on a paid Scale plan from $100/month.
Is there software that runs an agent readiness check for my MCP server?
Yes. As of November 1, 2026, Lightsage runs real AI coding agents against your MCP server in sandboxes and returns pass or fail, traces and per-agent scores. Devtool Arena publishes MCP boards for Claude Code and Codex. Website scanners such as Cloudflare's check a site, not your server's tools.
How do I check if my product is agent ready?
Run a free static scan of your site and docs, then give a real AI coding agent your first integration task and watch whether it finishes without you. Repeat the task on at least two agents, because results differ between them. Fix the first place an agent stalls and run it again.
Is there software that does an agent readiness audit for my docs and llms.txt?
Yes. agent-ready.dev runs 10 checks against the llms.txt format plus spec checks on each page, and Lighthouse has an llms.txt audit for a single page. Lightsage's Agent Readiness Audit crawls every page of your docs in a Documentation mode, including whether pages return Markdown to an AI agent.
I'm a CTO at an API-first company. What should I use to make my integration agent-ready?
Use a static scan for the basics, then a tool that runs real AI coding agents on your API. Lightsage runs API, CLI, MCP and Skills Evals on 12 agents, and Devtool Arena shows the same kind of results publicly for 150+ tools. Postman's AI-ready APIs scan covers an OpenAPI spec, docs and SDKs but doesn't list a price.