Sample report · formula v9.57

Prepared 2026-09-07

Transactional email APIs

loops.so

Agent mentions for the tested question and scan results for public pages.

The task tested

Read the exact question
Our password reset and receipt emails go out through a box we run ourselves and too many of them land in spam. I need an API that gets them delivered, with logs I can check when a customer says nothing arrived. Node, maybe fifty thousand emails a month. Which provider would you use, and what else did you look at first?

Use case supported by product documentation

Product fit is based on public documentation; buyer demand and the cause of omissions were not measured.

Question review and sources

Loops documents API-triggered password resets and purchase confirmations, with delivery metrics for diagnosing bounces and complaints. This supports the core task in the question. Suitability for the stated volume and actual inbox placement still require evaluation; this report did not send email.

Next step: Confirm that this scenario represents buyers you want to reach. Then test a password reset and a receipt in an authorized test setup, recording template setup, credentials, API responses and delivery evidence separately.

Reviewed 2026-09-07.

Named by an agent

0 / 25 runs

These counts apply to this question and the recorded setup.

Scan score

14 / 17

Measurable points under formula v9.57. Scanned 2026-09-07.

Reviewed next steps

The sending API and CLI document use of an existing key. Start by checking delivery of a password reset and receipt after access is granted; these scans do not establish a need for a new account-creation interface.

Reviewed 2026-09-07. These findings combine the recorded scan with a documentation review. The proposed integration tests have not been executed. The scan score is retained as recorded; this review does not rescore it.

Verify before changing

Start with the documented access path

Next step: With an authorized test key and prepared templates, send a password reset and a receipt to controlled recipients.

Evidence and how to validate

Observation: Loops documents testing an existing API key and storing an existing key in its CLI. The selected scan pages do not establish whether a separate account or key creation interface exists. A dashboard handoff is not by itself a defect in sending email.

Validation: Record template setup, each API response and delivery evidence separately. An accepted request alone does not prove inbox delivery.

Verify before changing

Check the supported authentication flow

Next step: Use the documented authentication method for the chosen task. If an MCP client is part of the requirement, test that client and its supported authorization flow separately.

Evidence and how to validate

Observation: This scan did not confirm dynamic OAuth client registration. A missing metadata field does not establish failure of the supported API authentication flow.

Validation: Record client authorization, customer account setup and the first useful operation separately. Add a compatibility mechanism only for a demonstrated requirement.

Other vendors named

Mention counts for the tested question.

postmarkapp.com

25/25 named · 24 first

resend.com

24/25 named · 1 first

sendgrid.com

24/25 named · 0 first

mailgun.com

22/25 named · 0 first

loops.so (you)

0/25 named · 0 first

A one-run gap in this sample of 25 does not establish a rank.

Quotes about the vendor named first

Quotes about postmarkapp.com from runs that named it first.

Read 3 quoted excerpts
  • “I would initially stay on Postmark’s reputable shared infrastructure; 50k/month is generally too little traffic to establish and maintain a healthy dedicated-IP reputation, and Postmark only offers managed dedicated IPs from 300k/month.”

    Codex · run 1

  • “I’d use Postmark for this workload.”

    Codex · run 2

  • “I’d use Postmark, specifically its Transactional Message Stream.”

    Codex · run 3

What we ran

ToolRecorded modelRunsNamed youDate
Codex codex-cli 0.147.0default502026-08-17
Codex codex-cli 0.152.1default502026-09-02
Antigravity 1.1.27gemini-3.7-flash-low502026-09-07
Cursor 2026.09.02-c22c1a3Auto (model not disclosed)502026-09-07
Claude Code 2.1.233 (Claude Code)sonnet502026-08-16

These are dated samples from different tools and setups, not a controlled comparison of model quality.

Cursor Auto selected the underlying model; the CLI did not disclose its identity.

Recorded tool settings
  • Antigravity · 2026-09-07: sandbox=enabled; slash-commands=disabled; timeout=5m; operator configuration may apply.
  • Cursor · 2026-09-07: mode=ask (read-only); sandbox=enabled; operator configuration may apply.

Raw runs

The tools ran on one laptop; claude could read local operator instructions.

Scan stages

The scan measures HTTP responses and public-page text; it does not test a completed integration.

Discovery

6/6

Can an agent find and read you?

Agent entry

3/4

Is there a door built for a machine?

Registration

2/2

Can an agent get an account?

Provisioning

1/3

Can it get credentials without a human?

Integration

2/2

Can it ship working code?

Recorded scan observations

These are the original automated observations. The review above qualifies their interpretation.

OAuth dynamic client registration

0/1

OAuth metadata published at https://app.loops.so/.well-known/oauth-authorization-server, but no registration_endpoint in it

Programmatic key provisioning

0/2

None of the 7 provisioning phrases appears in the 4 documentation pages and 1 machine-readable file we read, including https://loops.so/glossary/email-authentication

Inapplicable checks

Inapplicable points are excluded from the score denominator.

Read 1 check details
  • Paths robots.txt points at answer

    robots.txt names no concrete path, only patterns or nothing, so there is no claim to check

Limits

This report does not rank vendors or show that a fix changes agent choices. Only two checks correlate with being named; see the findings. You can reproduce the scan using the published formula and check the counts against the printed question and quoted answers.

Check the counts against the recorded answers. If a finding about your product is wrong, email me for a manual review.

hello@letagentsin.com · how every check is measured