Support caller-controlled aggregation budgets across client list APIs

未關閉
#1,996 1 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

維護者通常 1 天內回覆

還沒有人認領這個 Issue。

評估

難度
5/5
預估耗時
一週以上
新手友好度
35/100
Issue 類型
功能
描述清晰度
基本清楚
活躍度
活躍
技術堆疊
csharp

研究方向

Start by reviewing the auto-paginating client list helpers for tools, prompts, resources, resource templates, and Skills, plus the related pagination decisions in #1056 and #1993. Define a consistent opt-in budget and outcome for reaching it while preserving existing defaults, cancellation, opaque cursor behavior, and page-level caching. Done means tests cover finite and non-terminating pagination, page and entry limits, cancellation, and incomplete results.

由索引模型根據 Issue 內容生成。

描述

P3

Summary

Evaluate optional, consistent maximum-page and maximum-entry budgets for the auto-paginating client list APIs for tools, prompts, resources, resource templates, and Skills.

Single-page overloads already let applications control pagination themselves. This issue concerns convenience and robustness, not missing protocol pagination support.

Motivation

The convenience methods collect pages until the server ends pagination. A faulty or untrusted server can cause non-terminating requests or unbounded accumulated results; a legitimate server can also return more data than a caller wants to retain. Cancellation can bound elapsed time, but does not directly express page or entry budgets.

The historical empty-cursor reports in #1072 and #988 were closed by design: an empty string is a valid opaque cursor, and a server must omit or return null for nextCursor to finish. Optional caller budgets should protect callers without changing that interpretation.

Proposed scope

  • Evaluate a common opt-in page/entry budget design across the SDK's aggregating list helpers, preserving existing defaults, request options, and cancellation behavior.
  • Define budget counting at page boundaries, including empty pages with continuation cursors and pages larger than the remaining entry budget.
  • Define an explicit failure or clearly incomplete outcome when a budget is reached. Never silently present a truncated list as complete.
  • Consider repeated/non-terminating cursor behavior without parsing opaque cursors or treating an empty string as null.
  • Cover finite pagination, existing unbounded defaults, page and entry limits, cancellation, and non-terminating server responses in tests.

Boundaries

Keep entry-validation and rejected-entry reporting as a separate design topic. Coordinate their counting and incomplete-result semantics, but do not make this issue depend on a new validation API.

Keep page-level caching semantics intact; this does not require an invented aggregate TTL or an SDK-owned cache.

Related work

  • Companion: #1997, strict/diagnostic entry validation. Keep the budget and validation policies independent while coordinating counting and incomplete outcomes.
  • Tarek's follow-up comment on #1993.
  • #1072 and #988 provide the historical non-terminating pagination context; they should remain closed.
  • #1056 records the decision to expose explicit page control rather than duplicate the list helpers with Enumerate*Async APIs.
  • #1993 removes Skills-only aggregation limits in favor of evaluating a consistent SDK-wide approach.

[!NOTE]
This issue was drafted with GitHub Copilot.

主要語言
C#
星號
4.6k
分支
849
平均合併
1 天 16 小時
30 天內合併 PR
24

環境準備

在 Codespaces 中開啟

在瀏覽器裡用你自己的 GitHub 帳號啟動這個專案的開發容器。

  • 沒有 Dockerfile 或 Docker Compose 檔案
  • 沒有 Pull Request 範本
  • 閱讀貢獻指南

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

modelcontextprotocol/csharp-sdk 的其他 Issue

查看 modelcontextprotocol/csharp-sdk 的全部 Issue

相似的 Issue

更多 C# Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。