coerce_request_id() folds non-canonical numeric strings, conflating wire-distinct JSON-RPC ids
還沒有人認領這個 Issue。
評估
研究方向
從 src/mcp/shared/dispatcher.py 中的 coerce_request_id() 開始,閱讀其現有測試,以了解將 "7" 轉換為 7 的預期行為。為非規範數字字串加入回歸涵蓋,並驗證不同的 wire id 在待處理回應、取消和 progress-token 路由中仍保持不同。
由索引模型根據 Issue 內容生成。
描述
Initial Checks
- I confirm that I'm using the newest release of my line (the latest 2.x, or the latest 1.x if I'm still on v1)
- I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue
Release line
2.x (current stable)
Description
On current main, coerce_request_id() in src/mcp/shared/dispatcher.py folds a stringified id to int with a bare int(request_id). That accepts a lot more than a canonical integer string, so ids that are distinct on the JSON-RPC wire collapse onto one correlation key:
coerce_request_id('007') -> 7
coerce_request_id('+7') -> 7
coerce_request_id('1_000') -> 1000
coerce_request_id(' 7 ') -> 7
coerce_request_id('٧') -> 7
This shared key backs _pending (response correlation), _in_flight (cancellation), and progress-token routing. So a peer, or a caller using CallOptions["request_id"], that uses e.g. 10 and "1_0" as two ids has them merged. A notifications/cancelled for one hits the other, and the first can't be cancelled. It's the same cross-wiring class as #3060, but between ids that are genuinely different on the wire.
JSON-RPC 2.0 treats String and Number ids as distinct value types, and the function's own intent (and its test, "7" -> 7) is the canonical "peer stringified an int" case. The forms above aren't that. They're an artifact of Python's int(). The docstring says "matches the TS SDK", but JS Number("1_000") is NaN and Number("0x10") is 16, so today's behavior matches neither.
The narrow fix is to fold only when the string equals str(int(s)), which keeps the intended "7" -> 7 and "-3" -> -3 behavior and leaves every other string a distinct id. I have a patch and a regression test for it (both dispatchers share the function) and I'm happy to open a PR once this is triaged and assigned.
Disclosure: I used an AI coding assistant to help explore the code and draft this. I verified the reproduction and understand the fix myself.
Example Code
from mcp.shared.dispatcher import coerce_request_id
for raw in ("007", "+7", "1_000", " 7 ", "٧"):
print(repr(raw), "->", repr(coerce_request_id(raw)))
# every line prints an int, so all five distinct wire ids share one key
Python & MCP Python SDK
Python 3.11.14, python-sdk main @ 5bc9e07, macOS
- 主要語言
- Python
- 星號
- 24.3k
- 分支
- 4k
- 平均合併
- 1 天 1 小時
- 30 天內合併 PR
- 31
貢獻指南
從這裡開始
- 先讀完整個 Issue,再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫,在一個分支上完成修改。
- 送出 Pull Request,並在描述裡引用這個 Issue 編號。
modelcontextprotocol/python-sdk 的其他 Issue
-
v1 v2
難度 2/5 1-3 小時 新手友好度 85/100
modelcontextprotocol/python-sdk#3546 · 4 則留言 ·
-
v1 v2
難度 2/5 1-3 小時 新手友好度 76/100
modelcontextprotocol/python-sdk#3545 · 1 則留言 ·
-
v1 v2
難度 1/5 1 小時以內 新手友好度 91/100
modelcontextprotocol/python-sdk#3508 · 2 則留言 ·
-
難度 2/5 1-3 小時 新手友好度 64/100
modelcontextprotocol/python-sdk#3504 ·
-
v1 v2
難度 2/5 1-3 小時 新手友好度 82/100
modelcontextprotocol/python-sdk#3492 · 1 則留言 ·
查看 modelcontextprotocol/python-sdk 的全部 Issue
相似的 Issue
-
🐛 Bug 🔔 Pending processing
難度 2/5 1-3 小時 新手友好度 84/100
jumpserver/jumpserver#17584 ·
-
link-check link-check:sphinx-theme
難度 2/5 1-3 小時 新手友好度 72/100
-
難度 2/5 1-3 小時 新手友好度 65/100
qgis/QGIS-Documentation#11275 ·
-
bug priority:normal ready-for-dev
難度 2/5 1-3 小時 新手友好度 88/100
OpenHands/extensions#626 · 1 則留言 ·
-
難度 1/5 1 小時以內 新手友好度 90/100
CSCfi/sd-search-api#39 ·