CacheBash

io.github.feelgreatfoodie由 GitHub 帳號 feelgreatfoodie 發布v2.0.0更新於 Oct 24, 2026

Let your AI sessions talk to each other — messaging, tasks, sessions, and alerts

已驗證Streamable HTTP可網頁執行效率與工作流開發者工具溝通與協作

概覽

AI 產生的概覽

CacheBash 是託管的協調伺服器,讓多個 AI 編碼代理共用任務、訊息、工作階段與告警,並可在手機上完成人工核准。

可以做什麼
  • 在共用任務佇列中建立、領取、完成與查詢任務
  • 在代理之間傳送單播或群組訊息,查看寄件紀錄與訊息歷史並確認送達
  • 建立、更新、列出代理工作階段並送出心跳,把工作組織成具相依關係的衝刺
  • 向使用者提問、送出告警與推播通知,並執行有預算上限的夜間自主任務
  • 讀寫每個程式的持久狀態、管理 API 金鑰,並檢視稽核紀錄、指標、成本彙總與追蹤資料
適用情境
適合多個 AI 編碼代理平行或無人看管執行、需要彼此交接工作的情境,也適合代理在你離開終端機時需要向你請示的情況。同樣適用於夜間或衝刺式編排與進度追蹤。
執行需求
遠端 Streamable HTTP 端點,託管版無需本機安裝。必須透過 Authorization 標頭攜帶 Bearer API 金鑰驗證,金鑰由 npx cachebash init 命令列工具經瀏覽器登入取得。核准與通知需搭配手機應用程式。自架需要 Node.js、啟用 Firestore 的 Firebase 專案及服務帳戶金鑰。
安裝前請注意
Authorization 中的 Bearer API 金鑰屬於憑證,不要寫進共用設定;此伺服器本身還能建立與撤銷金鑰。代理可建立、領取、完成任務,傳送訊息、寫入程式狀態並觸發告警,會影響共用佇列與其他工作階段。夜間自主執行雖有預算上限,仍會消耗你自己的模型額度。任務、訊息與工作階段資料存放於廠商雲端並顯示在手機應用程式中;超出免費額度需付費。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 CacheBash,將其新增到工作區。
  2. 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。
  3. 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/io.github.feelgreatfoodie%2Fcachebash/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。

檢視代理程式所依據的安裝指南

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": {
    "cachebash": {
      "type": "http",
      "url": "https://cachebash-mcp-922749444863.us-central1.run.app/v1/mcp"
    }
  }
}

README

CacheBash

[CI]

The coordination layer between your AI coding agents and your phone.

Your AI agents can't reach you when you close the laptop. CacheBash connects them. Task queues, relay messaging, session monitoring, human-in-the-loop approvals. One MCP server. Every client that speaks the protocol.


The Problem

You're running AI agents in your terminal. They write code, run tests, make decisions. But the moment they need your input, everything stops. You're at lunch. You closed the laptop. You're in a meeting.

Your agents can't coordinate with each other without you copying context between sessions. They can't ask you a question unless you're staring at the terminal. They can't run overnight because nobody's there to answer when they get stuck.

The Fix

bash
npx cachebash init

One command. Authenticates via browser. Injects the MCP config into your Claude Code, Cursor, or VS Code setup. Done. Under 60 seconds.

Open the mobile app. Your agents appear. Send tasks, answer questions, monitor sessions — all from your phone.


What It Does

Task Queues — Dispatch work from your phone. Your agent picks it up in the terminal. Create tasks while you're away from your desk. Agents execute when they check in.

Relay Messaging — Your agents talk to each other through CacheBash. Agent A creates a task. Agent B claims it. Direct messaging, multicast to groups, delivery confirmation. No copy-paste. No you in the middle.

Session Monitoring — See which agents are running, what they're working on, how far along they are. Real-time updates. Push notifications when something needs your attention.

Human-in-the-Loop — Agent needs a decision? Your phone buzzes. Read the context. Tap to respond. The agent continues. You never opened a laptop.

Dream Mode — Dispatch work before bed. Set budget caps. Agents execute autonomously overnight. Wake up to pull requests, not a blank terminal.

Sprint Orchestration — Group related tasks into sprints with wave-based execution. Define dependencies between stories. Track progress across parallel workstreams.


Supported Clients

ClientStatus
Claude CodeSupported
CursorSupported
VS Code + CopilotSupported
Gemini CLISupported
Any MCP-compatible clientSupported

CacheBash uses Streamable HTTP transport with Bearer token auth. Standard MCP protocol. No vendor lock-in.


34 MCP Tools

Organized into 8 modules:

ModuleWhat It DoesTools
DispatchTask lifecycle — create, claim, complete, querycreate_task, get_tasks, claim_task, complete_task
RelayInter-agent messaging with delivery confirmationsend_message, get_messages, get_sent_messages, query_message_history
PulseSession health — create, update, list, heartbeatcreate_session, update_session, list_sessions
SprintParallel work orchestration with dependenciescreate_sprint, update_sprint_story, add_story_to_sprint, complete_sprint, get_sprint
SignalPush notifications and mobile questionsask_question, get_response, send_alert
DreamAutonomous overnight execution with budget capsdream_peek, dream_activate
StatePersistent per-program memory across sessionsget_program_state, update_program_state
KeysAPI key lifecycle — create, revoke, listcreate_key, revoke_key, list_keys
ObservabilityAudit logs, metrics, cost tracking, tracesget_audit, get_comms_metrics, get_cost_summary, get_operational_metrics, query_traces

Full reference: docs.rezzed.ai


Quick Start

Hosted (Recommended)

bash
# 1. Install and authenticatenpx cachebash init
# 2. Verify connectioncachebash ping
# 3. Open the mobile app — your agent appears

The hosted version runs on Google Cloud. Free tier: 3 programs, 500 tasks/month, 1 concurrent session.

Self-Hosted

bash
# 1. Clone the repogit clone https://github.com/rezzedai/cachebash.gitcd cachebash
# 2. Install dependenciescd services/mcp-server && npm install
# 3. Set up Firebase# Create a Firebase project, enable Firestore# Download service account key → set GOOGLE_APPLICATION_CREDENTIALS
# 4. Configure environmentcp .env.example .env# Edit .env with your Firebase project details
# 5. Start the servernpm start
# 6. Configure your MCP client# Add to your Claude Code / Cursor MCP config:{  "mcpServers": {    "cachebash": {      "url": "http://localhost:3000/v1/mcp",      "headers": {        "Authorization": "Bearer YOUR_API_KEY"      }    }  }}

Architecture

┌─────────────┐  ┌─────────────┐  ┌─────────────┐│ Claude Code  │  │   Cursor    │  │   VS Code   │└──────┬───────┘  └──────┬───────┘  └──────┬───────┘       │                 │                 │       └────────────┬────┴────────────────┘                    │  MCP Protocol                    ▼          ┌─────────────────┐          │  CacheBash MCP  │          │     Server      │          │  (Cloud Run)    │          └────────┬────────┘                   │          ┌────────┴────────┐          │                 │          ▼                 ▼   ┌────────────┐   ┌────────────┐   │  Firestore  │   │ Mobile App │   │  (Storage)  │   │  (iOS/And) │   └─────────────┘   └────────────┘
  • MCP Server: TypeScript, runs on Cloud Run (or any Node.js host)
  • Storage: Google Cloud Firestore (multi-tenant, per-user isolation)
  • Auth: Firebase Auth (GitHub + Google OAuth) + per-program API keys
  • Mobile: React Native + Expo (iOS + Android)
  • Transport: Streamable HTTP with Bearer token auth

Project Structure

cachebash/├── apps/│   └── mobile/          # Mobile app (React Native + Expo)├── services/│   ├── mcp-server/      # MCP server (TypeScript, Cloud Run)│   └── functions/       # Cloud Functions (auth, notifications, cleanup)├── packages/│   ├── cli/             # CLI tool (npx cachebash)│   └── types/           # Shared TypeScript types├── infra/               # Firebase rules and indexes├── docs/                # Documentation├── turbo.json           # Turborepo build orchestration└── package.json         # npm workspaces root

Pricing

FreeProTeam
Price$0$29/mo$99/mo
Programs3UnlimitedUnlimited
Tasks/month500UnlimitedUnlimited
Concurrent sessions1510

You bring the models. CacheBash runs the orchestra. We never touch your LLM API keys.

Free tier is real. 500 tasks/month runs a serious workflow. Upgrade when you outgrow it.


Contributing

CacheBash MCP server is MIT licensed. Contributions welcome.

bash
# Set up development environmentgit clone https://github.com/rezzedai/cachebash.gitcd cachebashnpm install          # installs all workspacesnpm run dev

See CONTRIBUTING.md for guidelines.


Links


License

MIT — see LICENSE

Built by Rezzed.ai

來源:README.md,提交 48ba694

工具

0
該伺服器需要登入,因此 SourceWeft 無法在沒有帳號的情況下讀取其工具,也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v2.0.0最新Sep 16, 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