Comparison · read on 13 October 2026
How AgenticWebScore compares
This page sets AgenticWebScore beside 7 other products that check whether a site is ready for agents. Each fact about them was read from their public pages on 13 October 2026, and every row names its page.
The comparison
A cell reads “Not stated” where the product’s own page does not give the fact.
| Product | Checks | Output | Method | API or MCP | Price and account | Source |
|---|---|---|---|---|---|---|
| AgenticWebScore | 61 checks in 5 dimensions, under rubric v0.7. | A score, a grade, a level, the evidence for each check, and a change and a command for each check to fix. | The rubric, the weights, the grade bands, and the level rules are public and versioned. Each record has a fingerprint. | A JSON API and an MCP server with 4 tools. | Free. No account. | Methodology |
| Cloudflare, isitagentready.com | 21 checks, mostly on draft standards. | A prompt to paste into a coding agent. The page says the prompt can contain errors. | Each check links to a standard. The page states no scale and no weights. | An MCP server with a scan_site tool, a URL Scanner API, and Radar data. | Not stated; the page shows no charge. No account. | isitagentready.com |
| Lighthouse, agentic browsing audits | About 4 areas: WebMCP, llms.txt, accessibility for agents, and layout shift (CLS). | Pass or fail for each audit. The page states no fix output. | Open source. The page states no formula and no weights. It runs on your own machine, so you can repeat a result. | None stated on the page. | Free. No account, except a trial program for WebMCP. | Chrome for Developers |
| GEO Optimizer (open source) | 9 categories, 27 user agents, and 47 content checks. | A geo fix command shows a preview, then writes files. | MIT licence. It runs on your own machine. | A command line, Python, a GitHub Action, and 12 MCP tools. | Free. A hosted plan starts at $19 a month. No account. | GitHub, Auriti-Labs/geo-optimizer-skill |
| GEO Toolbox | Access for 34 user agents under 6 browser identities, with pages rendered in a real browser. The page states no number of checks. | Fixes for the most serious problems. | The page publishes no rubric. | Not stated. | Free, with a limit for each IP address. It points to a paid platform. No account. | geotoolbox.ai |
| LLM Pulse | About 19 signals in 4 dimensions. | A prioritised list, each item linked to a standard. | It shows the evidence for each signal. The page states no weights. | None stated. | Free. A business email address is required. | llmpulse.ai |
| Frase | 10 checks. | A short guide. The full report needs a trial. | The page shows score bands and no method. | Not stated. | The score is free. The full report is a 7-day trial, with plans from $39 a month. The report needs a sign-up. | frase.io |
| Apify, Agent-Readiness actor | 5 signals with published weights. | A list of what is missing. The page states no fixes. | Weights and thresholds are published. It checks that a file exists, not how good it is. | An API and the Apify MCP server. | Priced by use; the page states no price. An Apify account and a token are needed. | Apify |
Where each fits
Use Cloudflare’s check or Lighthouse together with this site. They answer different questions, and a result that differs between them is a place to read the check page.
Cloudflare’s check links each of its checks to a standard and gives you a prompt for a coding agent. This site publishes its scale and its weights, so a score can be recomputed.
Lighthouse runs in a browser on one page at a time, and you can run it on your own machine. This site reads what a plain HTTP client receives, from the paths the rubric names.
What this site does not do
- It does not run JavaScript. It reads what a plain HTTP client receives, so a homepage whose HTML holds almost no readable text is reported as not measured.
- It does not monitor your site. A scan runs when somebody asks for one, and a stored record younger than 6 hours is reused. There are no alerts and no history chart.
- It does not fix anything. It gives the change to make and a command that repeats the measurement, and you or your agent writes the file.
What is different here
Each point below can be checked on this site.
- The rubric is public and versioned. The methodology lists all 61 checks, their weights, the grade bands, the level rules, and what changed in each version. The current version is v0.7.
- Every check has a command. Each page under /checks gives a shell command that repeats the measurement, so you can compare your result with the scanner’s.
- Every scan record has a fingerprint.
/verify/<fingerprint>recomputes it from the host, the rubric version, and the outcome of every check. It reports a mismatch as tamper evidence. - A site the scanner cannot read is not measured. It gets no score, not a low one, and the report says why.
- The API and the MCP server need no account.
/api/scanreturns the full scan record as JSON, described in OpenAPI. The MCP server at/mcphas 4 tools:scan_site,get_score,list_checks, andget_report.