QMesh MCP — 5 分鐘讓 Claude 變成你的 QA 助手

透過 Model Context Protocol 連接 Claude Code / Cursor / Claude Desktop / Codex

QMesh MCP 讓你的 AI 助理直接查詢 QMesh 平台的真實眾測資料：46 支去識別化 bug pattern、平台統計、測試員排行榜、方案報價。接 API Key 後還能讓 AI 主動把偵測到的問題送進 QMesh Signal Engine 由人類 QA 驗證、自動發包眾測、拉 bug 清單、產 release-readiness 報表（v0.5 新增）。

立即開始 →

🛠️ 提供的工具

工具	需要驗證	說明
`verify_install` v0.5.1	否	安裝後第一個試的工具：30 秒驗證 MCP 接通成功 + 提示下一步
`search_bug_patterns`	否	搜尋 46 支去識別化 bug pattern（含偵測 checklist、root cause 分類）
`get_platform_stats`	否	平台即時統計：測試員數、QA 認證數、企業數、bug / 任務數
`get_leaderboard`	否	日 / 週 / 月 / 年測試員排行榜，含 QIS 評分
`list_pricing_plans`	否	公開方案清單（預算、功能、退款政策）
`submit_ai_signal`	API Key	把 AI 偵測到的品質訊號送進 QMesh Signal Engine，由人類 QA 驗證
`create_test_task` v0.5	API Key	AI 自動發包眾測（免費方案：budget=0、限 5 個/24 小時）
`get_task_status` v0.5	API Key	追蹤任務狀態與 bug 統計
`list_bugs` v0.5	API Key	拉 bug 清單（可篩 severity / status，最多 100 筆）
`export_report` v0.5	API Key	產 Markdown release-readiness 報表

前 5 支工具不需登入即可使用（含 v0.5.1 新增 verify_install）；後 5 支需要 API Key。安裝後第一個試 verify_install，立刻知道有沒有接通成功。

🚀 快速安裝

選擇你正在使用的 AI 客戶端：

⚡ 一行指令安裝。Claude Code 是 Anthropic 的 CLI，原生 MCP 支援，最少 friction。

安裝 MCP server（read-only 模式）

claude mcp add qmesh -- npx -y @q-mesh/mcp

需要 submit_ai_signal 寫入功能（先到企業後台建 API Key）：

claude mcp add qmesh --env QMESH_API_KEY=qk_xxxxx -- npx -y @q-mesh/mcp

驗證

claude mcp list

應該看到 qmesh 在清單內。

在 Claude Code 中試

⚡ 最快驗證：「用 qmesh verify_install」 — 立刻回 ✅ 接通成功 + 提示下一步
「用 qmesh 找 critical 嚴重度的 security 類 bug pattern」
「幫我 review 這段 BaaS 程式碼有沒有踩 QMesh 的 pattern」

✅ 接通成功會回傳具體 pattern code（例 PAT-0004）+ 偵測 checklist。

打開 Cursor Settings → Features → MCP

點 Add new MCP server。

填入下列設定

Name：qmesh
Type：stdio
Command：npx -y @q-mesh/mcp

需 submit_ai_signal → 在 Environment variables 加上 QMESH_API_KEY=qk_xxxxx

重啟 Cursor，在 Composer / Chat 中試

「@qmesh 找 critical security pattern」

打開 Claude Desktop 設定檔

macOS：~/Library/Application Support/Claude/claude_desktop_config.json
Windows：%AppData%\Claude\claude_desktop_config.json

檔案不存在 → 自己新建一個。

貼這段（read-only 模式，無需 API Key）

{ "mcpServers": { "qmesh": { "command": "npx", "args": ["-y", "@q-mesh/mcp"] } } }

需 submit_ai_signal → 在 qmesh 區塊加 "env": { "QMESH_API_KEY": "qk_xxxxx" }

完全退出 Claude Desktop 再重開

Mac 用 Cmd+Q 或選單列「Quit Claude」；Windows 從工作列右下角退出。

⚠️ 「關閉視窗」不算退出。要徹底結束程式才會載入新的 MCP 設定。

開新對話試這幾句

「用 qmesh 找 critical 嚴重度的 security 類 bug pattern」
「QMesh 平台目前有多少測試員？」

⚠️ 需要 Codex CLI 含 MCP 支援的版本（2025+）。舊版請先升級。

打開 Codex 設定檔

路徑：~/.codex/config.toml（檔案不存在自行新建）

貼入 TOML 設定

[mcp_servers.qmesh] command = "npx" args = ["-y", "@q-mesh/mcp"] # 需 submit_ai_signal 才取消下面註解 # [mcp_servers.qmesh.env] # QMESH_API_KEY = "qk_xxxxx"

重啟 Codex CLI，查詢試試

「list qmesh tools」
「search qmesh patterns where category is security」

🔑 進階：開啟 submit_ai_signal（可寫入）

如果你是企業使用者、想讓 AI 把它發現的 bug 自動送到 QMesh：

登入 QMesh 企業後台
進入「API Keys」頁面 → 建立新 key（複製出來，只會顯示一次）
把剛才的 config 加上 env 區塊：

{ "mcpServers": { "qmesh": { "command": "npx", "args": ["-y", "@q-mesh/mcp"], "env": { "QMESH_API_KEY": "qk_你剛複製的 key" } } } }

重啟 Claude，丟下面這句驗證可寫入：

「用 qmesh submit_ai_signal 提交一個測試訊號到 task <你的 task_id>，title 是 'MCP onboarding test'」

✅ 成功會回傳 signal id（uuid 格式）。失敗請看下方 FAQ。

🧠 為什麼值得接 QMesh MCP

QMesh 的 bug pattern 來自真實眾測案例萃取，不是 LLM 生成的「best practice」。每支 pattern 都附：

真實情境描述（observed_behavior）
預期行為（expected_behavior）
偵測技術（detection_technique）
可驗證 checklist（≥ 3 條）

接 MCP 後，你的 AI 助理在生成 BaaS 程式碼前就能比對 pattern，等於把 QMesh 累積的踩雷紀錄當成 lint table 用。

🆘 常見問題

Claude 找不到 qmesh 工具，沒反應

通常是 config 沒生效。檢查：(1) 路徑要對，不是專案資料夾；(2) Claude Desktop 一定要 完全退出（Cmd+Q）再重開；(3) JSON 格式正確（沒多逗號、沒缺括號）。

看到 node: command not found

需要 Node ≥ 18。從 nodejs.org 安裝；裝完後 node --version 確認版本。

看到 invalid or expired api key

key 可能複製錯字、過期或被 revoke。回後台重新建一把（建議 90 天有效期），重啟 Claude。

看到 permission denied: task does not belong to api key owner

提交的 task_id 不是該 key owner 名下的任務。用你自己建立的 task_id，或請 task owner 建一把 key 給你。

看到 rate limit exceeded

submit_ai_signal 限制每 API key 每分鐘 10 次、每小時 200 次。等一下再試；正常使用下不會打到。

怎麼看 MCP 的 log 來 debug？

log 位置：~/Library/Logs/Claude/mcp-server-qmesh.log（Mac）。問題排不掉寄信 support@oneday.software 附上 log 末段。

🔗 相關連結

📦 npm 套件：@q-mesh/mcp
📚 Bug Pattern 索引：github.com/.../bug-patterns
🤖 給 LLM 的結構化說明：llms.txt
🏠 QMesh 首頁：q-mesh.com

QMesh MCP — Connect Claude in 30 Seconds

Model Context Protocol server for Claude Code, Cursor, Claude Desktop, Codex

QMesh MCP lets your AI assistant query real crowdtesting data: 46 de-identified bug patterns, live platform stats, tester leaderboard, pricing plans. With an API Key, AI can also submit detected quality signals to QMesh's Signal Engine for human QA verification, autonomously launch crowdtest tasks, pull bug lists, and generate release-readiness reports (added in v0.5).

Get Started →

🛠️ Available Tools

Tool	Auth	Description
`verify_install` v0.5.1	none	First tool to call after install: instant smoke test confirming MCP connection + suggested next prompts
`search_bug_patterns`	none	Search 46 de-identified bug patterns with detection checklist + root cause
`get_platform_stats`	none	Live metrics: testers, QA-certified, businesses, bugs, tasks
`get_leaderboard`	none	Day / week / month / year tester rankings with QIS
`list_pricing_plans`	none	Public plans with budget, features, refund policy
`submit_ai_signal`	API Key	Submit AI-detected signals into QMesh Signal Engine for human QA verification
`create_test_task` v0.5	API Key	AI launches a crowdtest task (free tier: budget=0, 5 tasks/24h limit)
`get_task_status` v0.5	API Key	Track task progress with bug stats by severity & status
`list_bugs` v0.5	API Key	Pull bug list (filter by severity/status, up to 100)
`export_report` v0.5	API Key	Generate Markdown release-readiness report

First 5 tools work without authentication (including v0.5.1 verify_install); the remaining 5 require an API Key. Try verify_install right after install — instant confirmation it's working.

🚀 Quick Start

Pick the AI client you're using:

⚡ One-line install. Claude Code is Anthropic's CLI with native MCP support — lowest friction.

Install MCP server (read-only mode)

claude mcp add qmesh -- npx -y @q-mesh/mcp

For submit_ai_signal write access (create an API Key in business dashboard first):

claude mcp add qmesh --env QMESH_API_KEY=qk_xxxxx -- npx -y @q-mesh/mcp

Verify

claude mcp list

You should see qmesh in the list.

Try it inside Claude Code

⚡ Fastest check: "use qmesh verify_install" — instant ✅ confirmation + next-step hints
"Use qmesh to find critical security bug patterns"
"Review my BaaS code against QMesh patterns"

✅ On success Claude returns concrete pattern codes (e.g. PAT-0004) with detection checklists.

Open Cursor Settings → Features → MCP

Click Add new MCP server.

Fill in

Name: qmesh
Type: stdio
Command: npx -y @q-mesh/mcp

For submit_ai_signal: add env var QMESH_API_KEY=qk_xxxxx

Restart Cursor, try in Composer / Chat

"@qmesh find critical security patterns"

Open Claude Desktop config

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %AppData%\Claude\claude_desktop_config.json

Create the file if it doesn't exist.

Paste this (read-only mode, no API Key needed)

{ "mcpServers": { "qmesh": { "command": "npx", "args": ["-y", "@q-mesh/mcp"] } } }

For submit_ai_signal: add "env": { "QMESH_API_KEY": "qk_xxxxx" } to the qmesh block.

Fully quit Claude Desktop and reopen

Cmd+Q on Mac or "Quit Claude"; quit from system tray on Windows.

⚠️ Closing the window is not enough — the app must fully exit.

Try in a new conversation

"Use qmesh to find critical security bug patterns"
"How many testers does QMesh have right now?"

⚠️ Requires Codex CLI with MCP support (2025+). Update if older.

Open Codex config

Path: ~/.codex/config.toml (create if missing)

Paste this TOML

[mcp_servers.qmesh] command = "npx" args = ["-y", "@q-mesh/mcp"] # Uncomment for submit_ai_signal: # [mcp_servers.qmesh.env] # QMESH_API_KEY = "qk_xxxxx"

Restart Codex CLI and try

"list qmesh tools"
"search qmesh patterns where category is security"

🔑 Advanced: enable submit_ai_signal (write access)

If you're a business user and want your AI agent to push detected bugs into QMesh:

Sign in to QMesh business dashboard
Go to "API Keys" → create a new key (copy it now, shown only once)
Add env block to the config:

{ "mcpServers": { "qmesh": { "command": "npx", "args": ["-y", "@q-mesh/mcp"], "env": { "QMESH_API_KEY": "qk_your_copied_key" } } } }

Restart Claude and verify with this prompt:

"Use qmesh submit_ai_signal to submit a test signal to task <your_task_id>, title 'MCP onboarding test'"

✅ Success returns a signal id (uuid). For failures see FAQ below.

🧠 Why connect QMesh MCP

QMesh bug patterns are extracted from real crowdtesting cases, not LLM-generated "best practices". Each pattern includes:

Real observed behavior
Expected behavior
Detection technique
Verifiable checklist (≥ 3 items)

Once connected, your AI assistant can compare against patterns before generating BaaS code — turning QMesh's accumulated incident records into a working lint table.

🆘 Troubleshooting

Claude doesn't see qmesh tools

Usually a config issue. Verify: (1) correct file path; (2) Claude Desktop must fully quit (Cmd+Q) and reopen; (3) JSON syntax is valid (no trailing commas, balanced braces).

node: command not found

Install Node ≥ 18 from nodejs.org. Verify with node --version.

invalid or expired api key

Key may be mistyped, expired, or revoked. Recreate one in the dashboard (90-day expiry recommended) and restart Claude.

permission denied: task does not belong to api key owner

You're submitting against a task_id not owned by the API key holder. Use your own task_id, or have the task owner issue you a key.

rate limit exceeded

submit_ai_signal is limited to 10/minute and 200/hour per key. Wait and retry; normal use won't hit this.

How to view MCP logs for debugging?

Mac log path: ~/Library/Logs/Claude/mcp-server-qmesh.log. Stuck? Email support@oneday.software with the log tail.

🔗 Resources

📦 npm package: @q-mesh/mcp
📚 Bug Pattern index: github.com/.../bug-patterns
🤖 LLM-readable summary: llms.txt
🏠 QMesh home: q-mesh.com