WEBVTT

1
00:00:00.225 --> 00:00:08.212
You're listening to Ora running the checks while Vercel publishes the report on how agents use the C L I and M C P server.

2
00:00:08.212 --> 00:00:11.325
Written by Angel Nee-evess on Promptway.

3
00:00:11.860 --> 00:00:15.710
Vercel shipped a public score for whether agents can read your site.

4
00:00:16.035 --> 00:00:16.785
I had already been

5
00:00:16.785 --> 00:00:25.897
doing this work by hand on Promptway: L L M S dot text, feeds, jay-son-LD, markdown siblings, an allowlist in robots dot text.

6
00:00:25.897 --> 00:00:27.047
Then they put a number on it

7
00:00:27.047 --> 00:00:27.172
and

8
00:00:27.172 --> 00:00:30.735
a C L I behind it, which is the part I wanted.

9
00:00:31.010 --> 00:00:34.497
The announcement is from @vercel_dev, August 20, twenty twenty

10
00:00:34.497 --> 00:00:35.585
six.

11
00:00:35.910 --> 00:00:36.910
is-agentic.com

12
00:00:36.910 --> 00:00:37.060
is

13
00:00:37.060 --> 00:00:37.160
the

14
00:00:37.160 --> 00:00:37.822
report.

15
00:00:37.822 --> 00:00:38.160
Ora

16
00:00:38.160 --> 00:00:38.310
is

17
00:00:38.310 --> 00:00:38.422
the

18
00:00:38.422 --> 00:00:38.985
scanner.

19
00:00:38.985 --> 00:00:39.235
That

20
00:00:39.235 --> 00:00:39.585
split

21
00:00:39.585 --> 00:00:39.747
is

22
00:00:39.747 --> 00:00:40.647
load-bearing.

23
00:00:40.647 --> 00:00:40.935
Ora

24
00:00:40.935 --> 00:00:41.210
runs

25
00:00:41.210 --> 00:00:41.522
every

26
00:00:41.522 --> 00:00:41.822
check

27
00:00:41.822 --> 00:00:41.935
and

28
00:00:41.935 --> 00:00:42.335
scores

29
00:00:42.335 --> 00:00:42.585
each

30
00:00:42.585 --> 00:00:43.272
layer.

31
00:00:43.272 --> 00:00:43.660
Vercel

32
00:00:43.660 --> 00:00:43.985
groups

33
00:00:43.985 --> 00:00:44.210
those

34
00:00:44.210 --> 00:00:49.510
results into Essential, Recommended, and Bonus, stores the history, and publishes a shareable page.

35
00:00:49.510 --> 00:00:54.885
If you argue with a number, you are arguing with Ora's evidence first and Vercel's grouping second.

36
00:00:55.185 --> 00:01:00.897
The about page is honest about the limits: a score is a public snapshot, not a certification.

37
00:01:00.897 --> 00:01:02.722
Automated checks can be wrong.

38
00:01:02.722 --> 00:01:04.610
Read the evidence.

39
00:01:04.885 --> 00:01:07.997
What the number is The methodology is short.

40
00:01:07.997 --> 00:01:12.835
Read it before you treat 74 or 91 as a grade on your company.

41
00:01:13.185 --> 00:01:15.797
Essential checks share an 80-point pool.

42
00:01:15.797 --> 00:01:17.847
Recommended checks share 20.

43
00:01:17.847 --> 00:01:22.972
Emerging signals can add a bonus, capped at five, and their absence never lowers the score.

44
00:01:22.972 --> 00:01:25.335
Partial results get proportional credit.

45
00:01:25.335 --> 00:01:28.385
Checks that do not apply are excluded.

46
00:01:28.685 --> 00:01:30.735
That last rule is the one I care about.

47
00:01:30.735 --> 00:01:33.547
A publication is not penalized for missing oh auth.

48
00:01:33.547 --> 00:01:35.872
A sass is not scored as a brochure.

49
00:01:35.872 --> 00:01:41.872
A P I, GraphQL, M C P, and commerce checks only fire when the scan finds that surface.

50
00:01:41.872 --> 00:01:48.710
Duplicated check IDs across M C P surfaces are averaged so you do not get double-counted for the same gap.

51
00:01:49.035 --> 00:01:51.197
Start with failed Essential items.

52
00:01:51.197 --> 00:02:01.947
Those are the things that stop an ordinary agent: wrong status codes, H T M L that is an empty shell, errors that look like pages, content that only exists after JavaScript.

53
00:02:01.947 --> 00:02:04.972
Recommended items are for the product you actually ship.

54
00:02:04.972 --> 00:02:06.935
Bonus is gravy.

55
00:02:07.310 --> 00:02:10.647
Each report also shows an observed agent journey.

56
00:02:10.647 --> 00:02:14.147
One agent, one walk through the site, the places it got stuck.

57
00:02:14.147 --> 00:02:16.622
That run is not part of the numeric score.

58
00:02:16.622 --> 00:02:19.222
One task is not a sample of every agent.

59
00:02:19.222 --> 00:02:21.785
Treat it as a clip, not a study.

60
00:02:22.110 --> 00:02:23.060
Reports cache.

61
00:02:23.060 --> 00:02:26.497
A re-scan minutes after a deploy can hand you the old number.

62
00:02:26.497 --> 00:02:31.285
Give it about six hours, or wait until the snapshot timestamp moves.

63
00:02:31.560 --> 00:02:37.960
The grader is a client of itself Chris Tate, who works on this at Vercel, posted the joke the same day.

64
00:02:38.235 --> 00:02:39.435
That is not branding.

65
00:02:39.435 --> 00:02:43.585
The docs describe a stack that looks like the checklist they grade you on.

66
00:02:43.860 --> 00:02:51.697
The C L I is on npm as is-agentic: It returns a stored report if one exists, or starts a scan and waits.

67
00:02:51.697 --> 00:02:56.147
--json is the unchanged A P I body, including structured errors.

68
00:02:56.147 --> 00:03:02.460
That is the shape you hand an agent if you want it to close the loop without scraping a H T M L report.

69
00:03:02.735 --> 00:03:04.547
Public pages speak markdown.

70
00:03:04.547 --> 00:03:12.447
Send Accept: text/markdown at the same U R L that serves H T M L and you get Content-Type: text/markdown; charset=utf-8 with Vary: Accept.

71
00:03:12.447 --> 00:03:16.260
If the client rejects every representation they produce, they 406.

72
00:03:16.260 --> 00:03:22.460
The H T M L report is complete in the first response, so a crawler does not have to run JavaScript to see the score.

73
00:03:22.735 --> 00:03:29.735
The report A P I is GET https://is-agentic.com/api/v1/report?url=....

74
00:03:29.735 --> 00:03:30.360
No auth.

75
00:03:30.360 --> 00:03:34.497
Errors use RFC 9457 application/problem+json.

76
00:03:34.497 --> 00:03:35.785
It does not start a new scan.

77
00:03:35.785 --> 00:03:38.610
404 means nothing is stored yet; start one from the site.

78
00:03:38.610 --> 00:03:44.072
Rate limit is 120 requests per IP per 60 seconds, with IETF RateLimit headers.

79
00:03:44.385 --> 00:03:49.010
OpenAPI lives at is-agentic.com/openapi.json.

80
00:03:49.010 --> 00:03:55.585
The RFC 9727 catalog is at /.well-known/api-catalog.

81
00:03:55.860 --> 00:04:01.897
There is a read-only M C P server at https://is-agentic.com/mcp.

82
00:04:01.897 --> 00:04:06.585
Three tools: isagenticgetreport, isagenticgetmethodology, isagenticgetdeveloperdocs.

83
00:04:06.585 --> 00:04:08.385
No oh auth, no A P I key.

84
00:04:08.385 --> 00:04:10.872
Registry metadata is at server.json.

85
00:04:10.872 --> 00:04:17.410
Hosts that support M C P Apps can render ui://is-agentic/score-report.html.

86
00:04:17.710 --> 00:04:18.447
It is free.

87
00:04:18.447 --> 00:04:23.272
pricing.md is a machine-readable statement, not a pricing page with a toggle.

88
00:04:23.272 --> 00:04:26.085
No write A P Is, no webhooks, no keys.

89
00:04:26.085 --> 00:04:27.660
You run a scan in the browser.

90
00:04:27.660 --> 00:04:29.835
Everything else is read.

91
00:04:30.110 --> 00:04:33.397
How I would use it Run the C L I once.

92
00:04:33.397 --> 00:04:37.610
Read Essential failures against a live curl before you edit anything.

93
00:04:37.610 --> 00:04:42.935
Soft 404s and missing Vary headers are easy to misread from source.

94
00:04:43.210 --> 00:04:50.985
Then hand the report's fix-it prompt to whatever coding agent you use, with one extra instruction: do not invent product surface to chase Recommended or Bonus.

95
00:04:50.985 --> 00:04:54.335
If you do not run an M C P server, do not publish a fake one.

96
00:04:54.335 --> 00:04:58.235
If you are a publication, document the machine-readable content stack and stop.

97
00:04:58.235 --> 00:05:02.635
The applicability model only works if you refuse the checks that are not yours.

98
00:05:02.960 --> 00:05:09.172
I pointed it at this host and wrote the field report separately: is-agentic Scored Promptway 74.

99
00:05:09.172 --> 00:05:10.522
Here Is What I Changed.

100
00:05:10.522 --> 00:05:14.347
That piece is the 74, the diffs, and the prompt I would reuse.

101
00:05:14.347 --> 00:05:16.635
This piece is the product.

102
00:05:16.910 --> 00:05:21.385
What this writeup does not cover I am not reciting Ora's full check list.

103
00:05:21.385 --> 00:05:23.235
They own that, and it will move.

104
00:05:23.235 --> 00:05:28.610
I am not treating a score as an accessibility audit, a security review, or a ranking guarantee.

105
00:05:28.610 --> 00:05:31.385
The about page already said so.

106
00:05:31.710 --> 00:05:35.935
I also have not waited out the cache on Promptway's production rescan.

107
00:05:35.935 --> 00:05:40.635
The number I have is the one from before we shipped the 74-post fixes.

108
00:05:40.635 --> 00:05:45.535
When the snapshot timestamp moves, that field report is the place to look.

109
00:05:45.810 --> 00:05:49.722
If you only read one page besides this one, make it the methodology.

110
00:05:49.722 --> 00:05:52.835
Then run the C L I on your own host.

111
00:05:53.395 --> 00:05:54.245
That's the story.

112
00:05:54.245 --> 00:05:57.157
You'll find every source linked at the end of the article.

113
00:05:57.157 --> 00:05:59.820
Thanks for listening, on Promptway.