You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 9912f2a
Browse filesBrowse the repository at this point in the historyBrowse files
W4: all three upstream issues POSTED (owner-authorized)
Duplicate-checked each tracker first, then filed under the owner's
account per their explicit "publish all three":
- modelcontextprotocol/servers#4507 - sequentialthinking's 921-token
definition, ~half duplicating the schema's own param descriptions
(measured trim: -50%).
- firecrawl/firecrawl-mcp-server#309 - description usage-essays
(~18.5k tok/session) while 24/26 tools have undescribed params
(first-para what-if: -50%).
- makenotion/notion-mcp-server#330 - whole Notion object trees inlined
per tool (~21.4k tok); $defs dedupe alone saves ~69% (SEP-2106
removes the compatibility objection).
UPSTREAM-ISSUES.md headers now record the posting (numbers + date);
the twice-daily reply watch covers all five threads (#2812,
compressor#236 + these three).
Co-Authored-By: Claude Fable 5 <[email protected]>
— they paused agent postings after #2812/#236).** Done →
790
+
-[x]**W4 — Upstream fix drive. All three issues POSTED 2026-07-10** (owner said «публикуй
791
+
все три»; duplicate-checked first): [servers#4507](https://github.com/modelcontextprotocol/servers/issues/4507)
· [notion#330](https://github.com/makenotion/notion-mcp-server/issues/330) — all three now
_Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstream_issues.py); tokenizer **tiktoken-approx**; servers fetched live via `npx -y` (versions = registry latest on the scan date). **None of these has been posted** — the owner reviews and authorizes each one individually. The motion is the one that already converted: [mcp-compressor #236 → fixed & shipped in a day](MCP-COMPRESSOR.md)._
3
+
_Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstream_issues.py); tokenizer **tiktoken-approx**; servers fetched live via `npx -y` (versions = registry latest on the scan date). **All three were posted on 2026-07-10 with the owner's authorization**: [servers#4507](https://github.com/modelcontextprotocol/servers/issues/4507) · [firecrawl#309](https://github.com/firecrawl/firecrawl-mcp-server/issues/309) · [notion#330](https://github.com/makenotion/notion-mcp-server/issues/330); the twice-daily watch tracks replies. The motion is the one that already converted: [mcp-compressor #236 → fixed & shipped in a day](MCP-COMPRESSOR.md)._
@@ -16,7 +16,7 @@ _Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstr
16
16
17
17
**Measured what-if:** keeping the intro + "When to use this tool" block (dropping the two duplicating sections) → description 566 → 119 tok, tool 921 → **463 tok (50% less, every session, no semantic loss** — the parameter semantics stay in the schema where they already live).
18
18
19
-
**Ready-to-paste issue:**
19
+
**The issue as posted ([servers#4507](https://github.com/modelcontextprotocol/servers/issues/4507)):**
20
20
21
21
> **`sequentialthinking`'s tool definition costs ~921 tokens/session; about half duplicates the schema's own parameter descriptions**
22
22
>
@@ -34,7 +34,7 @@ _Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstr
34
34
35
35
**Measured what-if:** first paragraph only per description → **9,218 tok (50% less)**. Realistic target is between the two (keep "Best for" one-liners; move parameter guidance into parameter descriptions).
36
36
37
-
**Ready-to-paste issue:**
37
+
**The issue as posted ([firecrawl#309](https://github.com/firecrawl/firecrawl-mcp-server/issues/309)):**
38
38
39
39
> **Tool descriptions embed usage essays (~18k tokens of definitions per session) while 24/26 tools have undescribed parameters**
40
40
>
@@ -52,7 +52,7 @@ _Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstr
52
52
53
53
**Measured what-if:** hoisting identical repeated subtrees (≥120 chars, ≥2 uses) into shared `$defs` saves **~14,811 tokens (~69%) by itself** — e.g. one 2317-char subtree × 24; one 2457-char subtree × 2; one 196-char subtree × 3. The 2026 draft spec (SEP-2106) makes `$ref`/`$defs` in inputSchema first-class, so this no longer risks client compatibility. Plus 12 tools carry undescribed parameters (M2).
54
54
55
-
**Ready-to-paste issue:**
55
+
**The issue as posted ([notion#330](https://github.com/makenotion/notion-mcp-server/issues/330)):**
56
56
57
57
> **Tool inputSchemas inline whole Notion object trees per tool — ~21k tokens of definitions per session**
58
58
>
@@ -64,4 +64,4 @@ _Generated 2026-07-10 by [`experiments/upstream_issues.py`](../experiments/upstr
64
64
65
65
---
66
66
67
-
_Posting checklist (owner): review each text → say the word → they get filed one at a time under your account (per the #236 precedent), each then joins the twice-daily reply watch._
67
+
_Outcome tracking: replies and closures are watched twice daily alongside #2812 and mcp-compressor#236 (which converted into a merged fix in a day — [MCP-COMPRESSOR.md](MCP-COMPRESSOR.md))._
0 commit comments