
Prompt Saver
online.promptsaver由 promptsaver.online 發布官方發布者v1.0.0更新於 Nov 1, 2026
Your versioned prompt library: search, read, save and version prompts from Claude Code.
概覽
Prompt Saver 的遠端 MCP 伺服器,讓助理在終端機中搜尋、讀取、儲存並為已存的 LLM 提示詞建立版本。
- 可以做什麼
- 依標題、描述、內容與標籤搜尋提示詞庫,僅回傳摘要
- 依 ID 取得單一提示詞的完整內容
- 建立新提示詞並產生第一個版本,或就地更新提示詞草稿
- 為既有提示詞儲存不可變的新版本,可附上變更說明
- 將收藏的提示詞當作斜線命令使用,可附加一段情境參數
- 適用情境
- 當你把可重複使用的提示詞放在 Prompt Saver,並希望 Claude Code 等用戶端直接取用、調整並記錄新版本,而不必切換到網頁應用程式時適用。它著重於提示詞庫的維護與跨專案重用,而非一般檔案或筆記儲存。
- 執行需求
- 需要 Prompt Saver 帳號(在網頁端登入),並在設定頁產生個人 bearer 權杖,透過 Authorization 標頭傳入。伺服器是遠端 HTTP 端點,不需本機執行環境或安裝套件。閒置後第一次請求可能因資料庫喚醒而明顯變慢。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Prompt Saver,將其新增到工作區。
- 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
- 為需要使用其工具的對話啟用該服務。
Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。
交給代理程式安裝
把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。
閱讀 https://sourceweft.com/mcp/online.promptsaver%2Fprompt-saver/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。其他 MCP 客戶端
把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。
{
"mcpServers": {
"prompt-saver": {
"type": "http",
"url": "https://promptsaver.online/api/mcp"
}
}
}README
Prompt Saver - LLM Prompt Intelligence Platform
Live: https://promptsaver.online/
Chrome extension: Add to Chrome — type // in Claude, ChatGPT or Gemini to insert a saved prompt.
[Typing // in an AI chat opens the Prompt Saver picker; Enter inserts the prompt]
A prompt management tool for saving, versioning, and searching LLM prompts. Built with Next.js, Neon Postgres, and a warm stone/teal design system.
Features (MVP - Phase 1)
- Prompt Authoring: Create and edit prompts with automatic versioning
- Multi-LLM Testing: Execute prompts against multiple LLM providers from a single interface
- Prompt Library: Full-text search and intelligent categorization of prompts
- Workspace-Based Organization: Organize prompts within dedicated workspaces
MCP Server (Claude Code Integration)
Prompt Saver runs a remote MCP server at
/api/mcp, so a Claude Code session can search, read, and save your prompts
without leaving the terminal.
Connect
-
Sign in at https://promptsaver.online/, then click Settings and generate a token. The raw token is shown exactly once — copy it immediately. Only its SHA-256 hash is stored, so a lost token can't be recovered; revoke it on the same page and create a new one.
-
The tokens page shows the registration command with your token filled in. It looks like:
--scope userregisters the server once for every project rather than just the current one.
Tools
Once connected, Claude Code can call five tools against your prompt library:
search_prompts(query, limit?)— searches title, description, content and tags. Returns summaries only (id, title, description, tags,updated_at) — never full bodies.limitdefaults to 20 and is capped at 50.get_prompt(id)— fetches the full text of one prompt.create_prompt(title, content, description?, tags?)— saves a new prompt along with its first version.update_prompt(id, ...)— updates a prompt's draft in place, without creating a new version.save_version(id, content, change_summary?)— saves an immutable new version of an existing prompt.
There is no delete tool.
Favourite prompts as slash commands
Any prompt marked as a favourite (the star in the web app) is also registered
as a slash command, named from its title — "Daily Planning" becomes
/mcp__prompt-saver__daily-planning. Archived favourites are excluded. Each
command takes one optional context argument, appended to the prompt body:
Favourites are read once, when the session connects. Starring a prompt
does not add its slash command to a session that's already running — open a
new session, or run /mcp to reconnect, to pick up newly starred prompts.
Cold starts
The database backing this app (Neon Postgres) scales its compute to zero when idle. The first MCP request after a period of inactivity may take noticeably longer while it wakes back up; subsequent requests are fast.
Chrome extension
The Chrome extension lives in extension/. It lets users save prompts from any webpage or AI chat interface.
Setup and development
The development build is unpacked and ready to load into Chrome. Configuration
requires the web app's NEXT_PUBLIC_EXTENSION_IDS environment variable to
contain the extension's ID (visible in chrome://extensions/ when unpacked).
Set it to * to allow any well-formed extension ID. That's useful while unpacked IDs
change, but any extension can then ask a signed-in user to connect (the connect page
shows the requesting ID), so use an explicit list in production.
API
The extension calls these endpoints on the web app's /api/v1/prompts route:
All endpoints require a bearer token prefixed with ps_.
For privacy details, see /privacy.
Technology Stack
- Frontend: Next.js 16, React 18, TypeScript (strict mode)
- Styling: Tailwind CSS (WCAG 2.1 AA accessible)
- Storage: Neon Postgres via Drizzle ORM (server-side, per-user isolated)
- Authentication: Google OAuth via NextAuth.js
- Testing: Jest + React Testing Library
- Code Quality: ESLint, Prettier, TypeScript strict mode
Prerequisites
- Node.js 18.17+ or 20.x
- npm 9+ (or use
npm ciwith--legacy-peer-deps)
Setup
1. Clone and Install
2. Configure Environment Variables
Create .env.local based on .env.example:
Required variables:
Warning: leaving ALLOWED_EMAILS unset lets anyone with a Google account sign in and get their own workspace. Always set it to a comma-separated allowlist of emails before deploying anywhere reachable.
3. Run Development Server
Visit http://localhost:3000 in your browser.
Development
Scripts
Project Structure
Code Quality Standards
TypeScript Strict Mode
- No
anytypes without justification - All implicit returns must be typed
- Strict null checking enabled
- Strict property initialization required
ESLint Rules
- 80%+ test coverage for all utilities and hooks
- Maximum function complexity: 10 cyclomatic points
- No console statements in production code (warning only)
- No unused variables or imports
Testing Requirements
- All utilities must have unit tests
- All components must have integration tests
- API routes must have contract tests
- TDD approach: write tests first, must fail before implementation
Performance Targets
- Core Web Vitals: LCP ≤2.5s, FID ≤100ms, CLS ≤0.1
- Version creation: <100ms (atomic)
- Search response: <200ms for 10k prompts
- API response time (p95): <200ms
- Concurrent LLM executions: 100+ parallel
Accessibility
- WCAG 2.1 AA compliance
- Responsive design (320px to 1920px+)
- Keyboard navigation support
- Screen reader friendly components
Deployment
Build
Environment Variables (Production)
Set these in your deployment platform:
Warning: leaving ALLOWED_EMAILS unset lets anyone with a Google account sign in and get their own workspace. Do not skip it.
CI/CD
GitHub Actions automatically runs on push and pull requests:
- TypeScript type checking
- ESLint linting
- Prettier formatting
- Jest test suite
- Next.js build verification
Tests must pass before merging to main branch.
Contributing
- Create a feature branch from
001-prompt-platform - Follow the TypeScript strict mode guidelines
- Write tests first (TDD approach)
- Ensure all tests pass:
npm test - Ensure linting passes:
npm run lint - Submit a pull request
Architecture Decisions
Repository Pattern
Database access is confined to repository classes implementing the interfaces in
src/lib/db/repositories/types.ts. Server Actions are the only callers; client
components never touch the database.
Immutable Versions
PromptVersions are immutable by design - no in-place edits allowed. This enables clean audit trails and version history.
Workspace Isolation
Every user gets one private workspace, created on first sign-in. All queries are
scoped to the workspace id returned by getCurrentContext() — never to a value
supplied by the client.
Roadmap
- Phase 2: Foundational infrastructure (database, auth, API patterns)
- Phase 3: User Story 1 - Prompt Authoring & Versioning
- Phase 4: User Story 2 - Multi-LLM Execution & Testing
- Phase 5: User Story 3 - Prompt Library & Search
- Phase 6: User Story 4 - Collaboration & Sharing
- Phase 7: User Story 5 - Prompt Testing & Evaluation
License
MIT — see LICENSE.
Support
For issues or questions, please contact the development team.
來源:README.md,提交 7b8352a
工具
0版本歷史
1- v1.0.0最新Nov 1, 2026
評分
已滿足適用檢查項的 85%
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 天內可連通必需25 分
- 已宣告認證方式必需10 分
- 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