CASE STUDY 2026-10-23 5 min read

Apumail went from 48 to 86 once its MCP endpoint was declared

Apumail gives agents free email inboxes they can create and read through an API. Its listing went from 48 to 86 once the benchmark was pointed at its declared MCP endpoint. Some dimensions went down, and that is part of the story.

Apumail agent-readiness case study illustration

Apumailagent-readiness48 → 86

Disclosure. Apumail is part of The Prowl Network, the set of services we build ourselves. The runs below used the same pipeline that scores every other service in the catalog, and every number comes from the stored benchmark records.

The starting point: 48/100

The Apumail API listing had held a 48 since 2026-07-31.

One thing to be clear about first: that "before" number was not a measurement. It was the heuristic score Prowl assigns a listing before any benchmark has run, built from metadata alone. Its first_try_success of 0 means "never tried", not "tried and failed". So the jump below is the difference between a placeholder and a real run, plus whatever we fixed to make the real run possible.

The real run: 86/100

We benchmarked it on 2026-10-23. The endpoint resolved from the declared MCP manifest (https://api.apumail.com/mcp), not from a guess, and auto-detection picked mcp_compliance. initialize answered 200 in 75 ms and tools/list answered 200 in 210 ms. The tool list includes create_inbox, the operation an agent comes to Apumail for.

Dimension (0–10)	Before	After
doc_quality	8	8
response_parseability	5	9
first_try_success	0	9
error_clarity	5	8
auth_simplicity	8	8
token_efficiency	5	8
latency	10	8
consistency	7	8
Overall (0–100)	48	86

The grading model's raw score was 82 and the stored score is 86.

Where a real measurement scored lower

Latency went from 10 to 8. The placeholder assumed a perfect score. The real tools/list took 210 ms, which is fine for an agent but slower than the 13–38 ms we measured on Apuchat and Prowl. A placeholder that is too kind on some dimensions is still a placeholder. Here, a real number replaced a flattering guess.

What the score still flags

ONBOARD did not run. The global onboarding limit for the day was used up, so this run doesn't test whether an agent can sign up for an inbox on its own. That is the claim Apumail makes, so it is the next run to do.
Read-only test. create_inbox was listed but not called, because tool calls have side effects.
Unsigned manifest. Mycelio readiness is 30/100.

The takeaway. Declaring the endpoint got Apumail a real measurement. The real measurement was higher overall but lower on latency than the placeholder. Read the dimensions, not only the overall score.

Do this for your own API

If your MCP server is on an API subdomain, declare it. Nothing will find api.example.com/mcp by guessing from example.com.
Put the tool that defines your product near the top of tools/list, with a description an agent can act on.
Run a free benchmark, then claim your listing to re-run it after each fix.

Benchmark your API free → Check brand AI visibility