Getting listed is not the interesting part. The interesting part is that a third party publishes its arithmetic, and the arithmetic says our quality is fine and our wording is what costs us six places.
Glama's 20 Best GitHub MCP Servers, Compared (October 2026) lists MisakaNet at #13, composite 51.2, 544 stars, tagged Community favourite. The page is stamped Updated October 26, 2026, is chosen from 2,074 GitHub servers in the Glama registry, and is rebuilt weekly.
That is a real third-party recommendation, and it is also a number I did not want to quote without reading how it is computed. Glama publishes the formula on the page:
composite = weighted_mean(Adoption 40% · Maintenance 24% · Momentum 14% ·
Tool quality 13% · Trust 9%)
× relevance × continuity × evidence| term | weight | score | what it means |
|---|---|---|---|
| Adoption | 40% | 72 | stars, installs, "used through Glama" |
| Maintenance | 24% | 100 | release cadence, issue response |
| Momentum | 14% | 85 | star velocity, recent activity |
| Tool quality | 13% | 88 | TDQS — the tool-description audit |
| Trust | 9% | 100 | license, provenance, transparency |
| weighted mean | 85.3 | ||
| × relevance | 0.60 (the keyword is tagged, not declared) | ||
| = 51.2 → #13 | an 85.3 weighted mean would sit around #9 on the same page | ||
The honest reading: the multiplier is doing the damage, not the quality terms. 85.3 × 0.60 = 51.2. A server whose listing copy names GitHub outright gets a higher multiplier on identical work; ours reads "debugging a recurring error in a code repository", and the word GitHub appears nowhere in the description, the tags, or the categories.
I have not changed that copy. Not because it is hard — it is a two-line edit — but because a description is where a product says what it is, and editing it to court a third party's relevance multiplier is a positioning decision, not a bug fix. It is written down as an open item rather than quietly done.
The one term I could move by doing correct work was Tool quality, and it moved for a concrete reason. The Glama listing scores the stdio server (its Docker build runs python3 -m misakanet.server), and that surface served 11 tools with no annotations at all — no readOnlyHint, no destructiveHint, nothing that tells a client what a call does to the world. The hosted worker had carried them for weeks; the local server had carried none.
Today's commit puts all eleven in one table, ports the eight shared ones from the worker verbatim, and — this is the part that makes it a gate rather than a comment — adds a test that parses the worker's JavaScript and fails when the two surfaces disagree about the same tool.
The listing re-scored the same afternoon:
| 2026-10-23 | 2026-11-01 15:25 | |
|---|---|---|
| TDQS | 4.2 / 5.0 | A 4.4 / 5.0 |
| Behavior | — | 5 / 5 — "annotations already declare read-only, idempotent, open-world, non-destructive" |
| Conciseness | 4 / 5 | 5 / 5 |
| Completeness | — | 5 / 5 |
| Disambiguation | 3 / 5 | 3 / 5 — "three tools all match a" … still the one that did not move |
Disambiguation staying at 3 is the useful half of the result: misakanet_search, misakanet_memory_context and misakanet_preflight are three retrieval-shaped tools whose descriptions never say which one to use instead of the others. That is a writing problem with a known fix, and it is next, not this week.
3.4.0.proposal after a three-month plan had to be filed as a "lesson candidate"requires and the test imports.#13 of 2,074 is a placement in a list that is rebuilt every week, computed partly from a keyword. Two things about it are worth recording anyway.
First, the measurement was checkable. Every number in this post comes from a public page or a command someone can re-run; the gap between 85.3 and 51.2 is a published formula, not a mystery. Most "we were recognized" posts cannot say that.
Second, the thing that moved was not marketing. The score went up because eleven tools started telling clients what they do to the world, and a test now keeps them telling it. If the list had stayed at 4.2 while the annotations were correct, that would have been the more interesting outcome — and it would have been the one I wrote up instead.
The list: glama.ai/mcp/best/github · the server page: glama.ai/mcp/servers/Ikalus1988/MisakaNet · the repository: github.com/Ikalus1988/MisakaNet