fix(cli): match @ MCP resources by any URL substring, and stop label wrapping - #5776
Closed
yiliang114 wants to merge 2 commits into
Closed
fix(cli): match @ MCP resources by any URL substring, and stop label wrapping#5776yiliang114 wants to merge 2 commits into
yiliang114 wants to merge 2 commits into
Conversation
…iption In reverse (`@`) completion mode, a suggestion row that carries a description (MCP server/resource entries) let the description column shrink the label box, so the label wrapped mid-string — the trailing `:` of `server:` landed on its own line. Pin the label (`flexShrink: 0`) when the row has a description so the description column absorbs the shrink and truncates instead. Description-less rows (plain files) keep the previous behaviour.
…ll servers Before this change, a bare `@<partial>` only matched MCP *server names* by prefix; a resource was reachable only after typing the full `@server:` prefix. Typing a fragment of a resource URI or name (e.g. `@analyze`) returned no match even when a resource URI contained it. Add cross-server resource discovery: at the bare `@<partial>` stage, match the partial case-insensitively as a substring against every resource's canonical `@server:uri` reference and its friendly name/title, across all configured servers, and surface those alongside files. Reuses the existing `ResourceRegistry.getAllResources()`; ranks prefix above substring and reference above name.
Contributor
There was a problem hiding this comment.
Pull request overview
This PR improves the CLI’s reverse (@) completion UX for MCP resources by (1) enabling cross-server substring matching of resources before a <server>: is typed, and (2) fixing reverse-mode suggestion row layout so labels don’t wrap mid-string when a description is present.
Changes:
- Add cross-server MCP resource “discovery” suggestions for bare
@<partial>using case-insensitive substring matching against<server>:<uri>and friendly name/title, mixed alongside file suggestions. - Adjust reverse-mode suggestion row flex behavior so description rows don’t squeeze/wrap the label.
- Add/extend unit tests covering resource discovery behavior and the reverse-mode label wrapping regression.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| packages/cli/src/ui/hooks/useAtCompletion.ts | Adds cross-server MCP resource discovery suggestions for bare @<partial> and integrates them into the search flow. |
| packages/cli/src/ui/hooks/useAtCompletion.test.ts | Adds tests for the new bare-@<partial> MCP resource discovery behavior (substring/name match, cross-server, files mixed in, trust gating). |
| packages/cli/src/ui/components/SuggestionsDisplay.tsx | Prevents reverse-mode labels from shrinking/wrapping when a row includes a description by pinning label shrink behavior. |
| packages/cli/src/ui/components/SuggestionsDisplay.test.tsx | Adds a regression test ensuring reverse-mode label + description stays on one line. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+181
to
+194
| return (registry.getAllResources?.() ?? []) | ||
| .map((resource) => { | ||
| const ref = buildMcpResourceRef(resource.serverName, resource.uri); | ||
| const friendly = resource.title || resource.name || ''; | ||
| return { | ||
| resource, | ||
| ref, | ||
| friendly, | ||
| rank: rankOf(ref.toLowerCase(), friendly.toLowerCase()), | ||
| }; | ||
| }) | ||
| .filter((m) => m.rank !== Infinity) | ||
| .sort((a, b) => a.rank - b.rank || a.ref.localeCompare(b.ref)) | ||
| .slice(0, MAX_SUGGESTIONS_TO_SHOW * 3) |
Comment on lines
+124
to
+130
| const output = lastFrame() ?? ''; | ||
| // The whole label (colon included) sits on a single line, before the desc. | ||
| expect(output).toContain('asys-mcp-http: MCP resource server'); | ||
| // And the label is not split across lines. | ||
| expect(output).not.toMatch(/asys-mcp-http\n/); | ||
| expect(output.split('\n').length).toBe(1); | ||
| }); |
Collaborator
Author
|
Closing as a duplicate of #5774, which opened earlier and is the more complete implementation (shared ranking helper + a minimum description-width floor). Extra test cases moved over as a comment there. |
Collaborator
|
Thanks for the PR, @yiliang114! Template looks good ✓ — all required sections filled in, bilingual, test plan with evidence. On direction: this is a clear-cut UX fix. Users remembering a fragment of a resource URI (the common case) getting "no match" is a real usability gap. Extending On approach: the scope is tight — exactly two focused fixes, +272/-12 (mostly tests), 4 files, no drive-by refactors or scope creep. The ranking scheme (ref prefix > name prefix > ref substring > name substring) is sensible and produces intuitive ordering. The bare- Moving on to code review. 🔍 中文说明感谢 @yiliang114 的 PR！ 模板完整 ✓ — 所有章节齐全，双语，测试计划附带证据。 方向：这是一个明确的 UX 修复。用户只记得资源 URI 的片段（最常见的场景）却得到"无匹配"，是真实的可用性问题。将 方案：范围精确——恰好两个聚焦的修复，+272/-12（主要是测试），4 个文件，没有顺手重构或范围膨胀。排序机制（引用前缀 > 名称前缀 > 引用子串 > 名称子串）合理且产生直觉排序。空 进入代码审查 🔍 — Qwen Code · qwen3.7-max |
Collaborator
Code ReviewClean implementation. I read the diff against my own proposal (add a A few things worth noting:
No correctness bugs, security issues, or AGENTS.md violations found. TestingUnit TestsAll 38 tests pass across the two affected suites: New test coverage includes: substring matching across servers, name/title matching, mixed-with-files, empty- Build + Headless SmokeTmux Interactive TestLaunched the CLI interactively in tmux at The CLI starts cleanly and renders the TUI. Note: 中文说明代码审查实现干净。我将 diff 与自己的方案（添加一个遍历 值得注意的几点：
未发现正确性 bug、安全问题或 AGENTS.md 违规。 测试所有 38 个测试通过（33 个 — Qwen Code · qwen3.7-max |
Collaborator
|
This is a good PR and I'm confident recommending it for merge. Going back to my independent proposal: I would have done essentially the same thing — add a substring-matching function over The code is straightforward — no over-engineering, no speculative abstractions. The The test suite is strong: 6 new unit tests covering substring matching, cross-server discovery, name/title matching, mixed-with-files, empty-trigger guard, and untrusted-folder guard, plus the layout regression test. The build is clean and the CLI works. The only limitation in my verification is that I couldn't exercise MCP resource completion end-to-end in the TUI (no MCP server in CI), but the 33 unit tests covering all MCP discovery paths give high confidence. Approving. ✅ 中文说明这是一个好的 PR，我很有信心推荐合并。 回到我的独立方案：我本质上会做同样的事——在 代码直截了当——没有过度工程化，没有投机性抽象。 测试套件扎实：6 个新单测覆盖子串匹配、跨 server 发现、名称/标题匹配、与文件混排、空触发守卫和不信任目录守卫，加上布局回归测试。构建干净，CLI 正常工作。 验证的唯一限制是无法在 CI 的 TUI 中端到端验证 MCP 资源补全（无 MCP server），但 33 个覆盖所有 MCP 发现路径的单测提供了充分信心。 批准合并 ✅ — Qwen Code · qwen3.7-max |
qwen-code-ci-bot
approved these changes
Jun 23, 2026
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
LGTM, looks ready to ship. ✅
Contributor
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this PR does
Two fixes to
@completion for MCP resources, both surfaced by #5733.The main behavior change: typing a bare
@<fragment>now matches MCP resources by any substring of their@server:urireference or friendly name, across all configured servers — not just server names by prefix. So@analyze,@asight,@bubbleall reach a resource whose URI contains that fragment, mixed in with file results, without first typing the full@server:prefix.The second fix is layout: in
@mode a suggestion row that has a description (MCP server/resource entries) no longer wraps its label mid-string.Why it's needed
#5733 added "match MCP resource completions by name and discover servers", but discovery before the colon only matched server names by prefix. A user who remembers a fragment of a resource URI (the common case) got nothing:
@analyzereturned "no match" even thoughasys-mcp-http:asight://skills/analyze_ppu_opexists. Reaching a resource required knowing and typing the exact server name first.Separately, #5733 was the first time
@suggestions carried a description. That exposed a layout bug: the description column claimed nearly the full row width, squeezing the label so it wrapped — the:of aserver:entry rendered on its own line.Reviewer Test Plan
How to verify
Configure any MCP server that exposes resources whose URIs contain a recognizable mid-string (e.g.
asight://skills/analyze_ppu_op), then in the TUI:@analyze— Before: "no match". After: the matching resources appear (asys-mcp-http:asight://skills/analyze_ppu_op,…/analyze_ppu_bubble).@asys-mcp— Before: theasys-mcp-http:server entry's:wraps to the next line. After: label + description on one line, resources also listed directly.@asight(scheme),@skill(mid-URI, mixes with.qwen/skills/files),@bubble(tail),@schema(a differentdb://resource).@server:urireference.Evidence (Before & After)
Verified live in tmux against a real stdio MCP server (
asys-mcp-http,/mcpshows ✓ connected), 120-col terminal.Before (from the original report):
@asys-mcp→ server entry rendered with the:wrapped onto a second line.@analyze→无匹配/ no match.After:
Unit coverage:
useAtCompletion.test.ts(substring across servers, name/title match, mixed-with-files, empty-@stays files-only, untrusted folder) andSuggestionsDisplay.test.tsx(reverse-mode label stays on one line). 38 passing across the two suites; lint + typecheck clean.Tested on
Environment (optional)
Local build (
npm run build), runningpackages/cli/dist/index.jsagainst a stdio MCP server. Windows/Linux not exercised locally — left to CI.Risk & Scope
@keystroke. It is synchronous over the in-memoryResourceRegistryand capped atMAX_SUGGESTIONS_TO_SHOW * 3, so it is cheap; very large registries would do more substring work per keystroke.getMcpResourceSuggestionsbehavior and avoid a new dependency. A very long reference in a very narrow terminal may overflow rather than wrap (mirrors the prior long-label truncation path).@stays files-only; untrusted folders still surface no resources; the@server:drill-in path is unchanged.Linked Issues
No linked issue — reported directly. Follows up on #5733, which introduced
@MCP resource completion.中文说明
这个 PR 做了什么
针对 MCP resource 的
@补全修两个由 #5733 暴露的问题。主要行为变化：现在输入裸
@<片段>，会用该片段去子串匹配所有已配置 server 的资源（匹配其@server:uri全引用或友好名），而不再只按前缀匹配 server 名。于是@analyze、@asight、@bubble都能匹配到 URI 含该片段的资源，并与文件结果混排，不需要先打出完整的@server:前缀。第二个修复是布局：
@模式下带 description 的建议行（MCP server/resource 项）不再把标签从中间折行。为什么需要
#5733 加了「按名匹配 MCP 资源补全并发现 server」，但冒号之前的发现只按前缀匹配 server 名字。只记得资源 URI 片段的用户（最常见的情况）什么都得不到：
@analyze返回「无匹配」，尽管asys-mcp-http:asight://skills/analyze_ppu_op是存在的。要够到资源必须先知道并打出准确的 server 名。另外，#5733 是
@建议第一次带 description，由此暴露一个布局 bug：description 列几乎吃满整行宽度，把标签挤到折行 ——server:项的:被甩到了下一行。Reviewer Test Plan
如何验证
配置任意一个暴露资源、且资源 URI 含可辨识中段（如
asight://skills/analyze_ppu_op）的 MCP server，然后在 TUI 中：@analyze—— 修复前：「无匹配」。**修复后：**命中相关资源。@asys-mcp—— 修复前：asys-mcp-http:server 项的:折到下一行。修复后： 标签和描述同一行，资源也直接列出。@asight（scheme）、@skill（URI 中段，且与.qwen/skills/文件混排）、@bubble（末尾）、@schema（另一个db://资源）。@server:uri引用。证据（前 / 后）
在真实 stdio MCP server（
asys-mcp-http，/mcp显示 ✓ 已连接）、120 列终端下 tmux 实测。英文正文中的终端输出即为「修复后」实况；「修复前」为原始报告中的两种坏状态（冒号折行、@analyze无匹配）。单测覆盖：
useAtCompletion.test.ts（跨 server 子串、名/标题匹配、与文件混排、空@仍只出文件、不信任目录不出资源）与SuggestionsDisplay.test.tsx（reverse 模式标签不折行）。两套件 38 项通过；lint + typecheck 干净。风险与范围
@按键时执行，但它是对内存中ResourceRegistry的同步遍历，并按MAX_SUGGESTIONS_TO_SHOW * 3截断，开销很小。getMcpResourceSuggestions行为一致，且不引入新依赖。极窄终端下的超长引用可能溢出而非折行（沿用原有长标签截断路径）。@仍只出文件；不信任目录仍不出资源；@server:下钻路径不变。