
Supstack
io.github.DrBaher由 GitHub 帳號 DrBaher 發布v0.18.2更新於 Oct 27, 2026
Evidence-based supplement research, dosing, safety, and interactions as agent-callable tools.
概覽
Supstack 將循證補充品研究、劑量、安全性與交互作用查詢包裝成 MCP 工具,資料來自公開的 SupStack API。
- 可以做什麼
- 依名稱或目標搜尋、研究補充品,並對兩到三種做並列比較
- 查詢研究文獻、交互作用路徑、補充品與藥物的交互作用,以及術語釋義
- 管理本機補充品清單(含劑量、服用時間、品牌),並依目標涵蓋度給出 A–F 評等
- 匯出清單;登入後還可取得個人化推薦、檢視或設定健康檔案、列出 N-of-1 實驗、記錄服用與遵從度
- 適用情境
- 適合需要助理以證據回答補充品問題、在搭配補充品或藥物前檢查交互作用,或協助維護並評估個人補充品清單與服用紀錄的情境。
- 執行需求
- 以本機 stdio 程序執行,來自 npm 套件 @supstack/cli(或 Homebrew),透過 mcp 子指令啟動。需要連線至 SupStack API 的網路;匿名讀取不需帳號。帳號相關工具需透過 login 裝置碼流程登入或提供 SUPSTACK_TOKEN。選用設定包括 SUPSTACK_API_KEY、SUPSTACK_API_URL 與 SUPSTACK_HOME。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Supstack,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
@supstack/cli
Evidence-based supplement intelligence in your terminal — and an MCP server for AI agents.
A thin client over the public SupStack API. Read-only, no account required. One capability registry powers both the CLI and the MCP server.
Status: Phase 2 (accounts) in progress. All read commands below work, plus an MCP server and
login/whoami. Install withnpm install -g @supstack/cli.
Install
Then enable shell completion: supstack completion install.
From source (in this directory): npm install && npm run build && node dist/index.js define adaptogen.
Usage
Account (Phase 2)
login shows a one-time code, opens https://supstack.me/activate, and finishes
once you approve in the browser. The token lives in ~/.supstack/config.json
(0600). Reads work fully anonymously — an account unlocks personalized features.
Once logged in, sync your stack with your account:
As an MCP server
This gives an agent the full SupStack toolset. Read-only (no account needed):
supstack_research, supstack_search, supstack_compare, supstack_studies,
supstack_interactions, supstack_stack, supstack_rate_stack, supstack_export,
supstack_define.
Account-scoped tools are also exposed — supstack_recommend,
supstack_profile_get / supstack_profile_set, supstack_experiments_list /
supstack_experiments_get, supstack_track_log, supstack_track_adherence
(plus cloud supstack_stack pull/push/sync). These require the user to be signed
in (supstack login, or a SUPSTACK_TOKEN); without a token they return a clear
"not logged in" error rather than failing. Mutating tools (profile_set,
track_log) are flagged readOnlyHint: false for the agent.
Claude Code (one command):
Claude Desktop — add to claude_desktop_config.json
(macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):
Or run it without a global install via
"command": "npx", "args": ["-y", "@supstack/cli", "mcp"]. To use a local checkout, point at the built entrypoint —"command": "node", "args": ["/absolute/path/to/supstack-cli/dist/index.js", "mcp"](runnpm run buildfirst).
Configuration
Global flags available on any command: --json, --no-cache, --timeout <seconds>, --color / --no-color, -q, --quiet.
supstack auth set-key <key> persists a key to ~/.supstack/config.json (written 0600).
Response cache
Read-only API responses are cached under ~/.supstack/cache/ (1-hour TTL by
default) to keep repeat lookups well under the 60/min rate limit. Bypass it per
command with --no-cache, or manage it with supstack cache clear /
supstack cache path. The cache is bounded (oldest entries pruned past a cap).
Shell completion
One-step install for your current shell (writes the script and wires your rc file
idempotently; uninstall reverses it):
…or generate the script yourself:
Completions are dynamic — the script forwards what you've typed to
supstack __complete, which offers the right thing for the position:
- top-level commands and sub-actions (
stack <TAB>→add remove list pull push sync) - supplement slugs where a slug is expected (
research <TAB>,compare a <TAB>,stack add <TAB>,track log <TAB>) - goal ids after
search --goal <TAB>
Slug/goal lists are fetched once from the API and cached under
~/.supstack/completion/ (24-hour TTL). Pre-warm or refresh them with:
Exit codes
Commands exit with a semantic code so scripts and MCP wrappers can branch on the kind of failure:
Develop
See CLAUDE.md for the capability pattern and how to add a command.
License
MIT
來源:README.md,提交 03d5516
工具
0版本歷史
1- v0.18.2最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分