Lighthouse Agentic Browsing category (Chrome)
Audit or test tool operated by Google Chrome (Lighthouse).
In short
An experimental Lighthouse category, run in Chrome 150 or later, that audits a page for machine interaction with deterministic checks and reports a pass ratio instead of a 0 to 100 score. 6 sourced facts follow, each with its read date.
What it is
An experimental Lighthouse category, run in Chrome 150 or later, that audits a page for machine interaction with deterministic checks and reports a pass ratio instead of a 0 to 100 score.
Operator: Google Chrome (Lighthouse). Type: Audit or test tool. Home page: developer.chrome.com/docs/lighthouse/agentic-browsing.
Ten-criterion rubric
| Criterion | Points | What the source says |
|---|---|---|
| Journey: find | 4 of 4 | ARD manifest audit 1 |
| Journey: read | 4 of 4 | llms.txt and accessibility tree audits 1 |
| Journey: use | 4 of 4 | WebMCP tool and schema audits, layout stability 1 |
| Journey: pay | 0 of 3 | no payment audits 1 |
| Checks | 2 of 15 | 7 audits are listed for the category (three WebMCP, llms.txt, ard.json, accessibility for agents, layout stability); under the 10-check band 1 |
| Real agents | 0 of 15 | deterministic audits, no agent run 1 |
| Open method: catalog | 3 of 3 | each audit has a documentation page 1 |
| Open method: weights | 0 of 3 | no weighted score by design 1 |
| Open method: scale | 0 of 2 | pass ratio, no grade scale 1 |
| Open method: changes | 2 of 2 | scoring page dated 5 May 2026 1 |
| Standards: cites | 1 of 3 | 2 of its 7 agentic-browsing audits link the spec they test: the ard.json audit page links the ARD spec and the llms.txt audit page links llmstxt.org; the three WebMCP audit pages link none, and the two audits on accessibility and layout stability link web.dev articles, not specs (29 percent) 2 |
| Standards: maintains | 2 of 2 | Google is named on agentready.org as a collaborator on the AgentReady standard 3 |
| Dev fit: api | 3 of 3 | runs as a Node module for programmatic use; there is no hosted API 4 |
| Dev fit: mcp | 0 of 3 | no MCP server or tool is named on the Lighthouse pages; WebMCP is an audit target 1 |
| Dev fit: cli | 2 of 2 | Lighthouse CLI 1 |
| Dev fit: ci | 2 of 2 | called suitable for CI/CD 1 |
| Peers: ranking | 0 of 6 | none |
| Peers: size | 0 of 4 | none |
| Fixes: guidance | 4 of 4 | each audit page explains the fix 5 |
| Fixes: brief | 0 of 3 | each audit page explains its own fix; no single brief or prompt 1 |
| Fixes: applies | 0 of 3 | none |
| Repeat scans | 1 of 5 | calls the audits deterministic and lists causes of fluctuation; no data 1 |
| Free: free | 3 of 3 | free and open source 1 |
| Free: limits | 2 of 2 | no cap 1 |
Total 39 of 100, rank 7 of 10. The ranking is on the pick page; the same cells are on the evidence page.
What the sources say
- The Agentic Browsing category evaluates how well a site is constructed for machine interaction through a set of deterministic audits.6
- Unlike other Lighthouse categories it has no weighted average score from 0 to 100; the report shows a fractional score (how many agentic readiness checks pass), pass or fail status for specific audits and informational counts.6
- The category and its WebMCP support are marked experimental and based on proposed standards; testing it requires Chrome 150 or later, and WebMCP audits require registering for the WebMCP origin trial.6
- Its audits cover WebMCP tool registration (declarative and imperative tools, read through the Chrome DevTools Protocol WebMCP domain), a subset of accessibility audits (names and labels, tree integrity, visibility), Cumulative Layout Shift, the presence of llms.txt at the domain root, and an ard.json manifest.6
- The ard.json audit looks for an Agentic Resource Discovery manifest through an Agentmap directive in robots.txt, a rel="ard" link, a Link header or /.well-known/ard.json, falls back with a warning to ai-catalog.json, and validates it against a JavaScript port of the ARD v0.91 conformance suite without fetching the artifacts it points to.2
- The documentation says results can fluctuate when tools are registered dynamically in JavaScript, when the accessibility tree changes, or when layout shifts move elements; it calls the audits suitable for CI/CD pipelines.6
What it does not do
- Does not give a weighted 0 to 100 score or a ranking; the documentation says the focus is signals, not a definitive ranking.
- Does not run an AI agent on a task; its audits are deterministic checks of the page.
Where to read next
Shown next to Ora because both look at llms.txt and WebMCP presence; Lighthouse reports a pass ratio for a page in Chrome, Ora a 0 to 100 score for a domain.
Sources
- developer.chrome.com/docs/lighthouse/agentic-browsing/scoring. https://developer.chrome.com/docs/lighthouse/agentic-browsing/scoring, read 2026-11-05.
- developer.chrome.com/docs/lighthouse/agentic-browsing/ard. https://developer.chrome.com/docs/lighthouse/agentic-browsing/ard, read 2026-11-05.
- agentready.org/llms.txt. https://agentready.org/llms.txt, read 2026-11-05.
- developer.chrome.com/docs/lighthouse/overview. https://developer.chrome.com/docs/lighthouse/overview, read 2026-11-05.
- developer.chrome.com/docs/lighthouse/agentic-browsing/webmcp-schema-validity. https://developer.chrome.com/docs/lighthouse/agentic-browsing/webmcp-schema-validity, read 2026-11-05.
- Lighthouse Agentic Browsing docs. https://developer.chrome.com/docs/lighthouse/agentic-browsing, read 2026-11-05.
Last reviewed 2026-11-05