mcp-shrimp-task-manager MCP — Сценарии использования, Установка & Живое демо

Зачем использовать

Ключевые функции

Natural language → structured tasks with dependencies
Continuous mode runs tasks sequentially without re-prompting
Research mode for exploration before commit
Per-project rules enforced across agent turns

Живое демо

Как выглядит на практике

shrimp-task-manager.replay ▶ готово

0/0

Установка

Выберите клиент

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-shrimp-task-manager"
      ],
      "_inferred": true
    }
  }
}

Откройте Claude Desktop → Settings → Developer → Edit Config. Перезапустите после сохранения.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-shrimp-task-manager"
      ],
      "_inferred": true
    }
  }
}

Cursor использует ту же схему mcpServers, что и Claude Desktop. Конфиг проекта приоритетнее глобального.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-shrimp-task-manager"
      ],
      "_inferred": true
    }
  }
}

Щёлкните значок MCP Servers на боковой панели Cline, затем "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-shrimp-task-manager"
      ],
      "_inferred": true
    }
  }
}

Тот же формат, что и Claude Desktop. Перезапустите Windsurf для применения.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "shrimp-task-manager",
      "command": "npx",
      "args": [
        "-y",
        "mcp-shrimp-task-manager"
      ]
    }
  ]
}

Continue использует массив объектов серверов, а не map.

~/.config/zed/settings.json

{
  "context_servers": {
    "shrimp-task-manager": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-shrimp-task-manager"
        ]
      }
    }
  }
}

Добавьте в context_servers. Zed перезагружается автоматически.

claude mcp add shrimp-task-manager -- npx -y mcp-shrimp-task-manager

Однострочная команда. Проверить: claude mcp list. Удалить: claude mcp remove.

Сценарии использования

Реальные сценарии: mcp-shrimp-task-manager

Ship a multi-day feature across Claude sessions with Shrimp

👤 Solo devs on longer features ⏱ ~45 min intermediate

Когда использовать: A feature is bigger than one Claude context window — you don't want to lose state.

Предварительные требования

Repo cloned + built — npm install && npm run build in mcp-shrimp-task-manager; point .mcp.json at the data dir

Поток

Plan

Use Shrimp plan_task on 'implement two-factor auth'. Produce tasks with dependencies.✓ Скопировано

→ Task graph saved to disk
Execute one task at a time

execute_task next. When done, mark complete and show me the next.✓ Скопировано

→ One task finished, next queued
Resume tomorrow

(new session) Read Shrimp state and continue from the first open task.✓ Скопировано

→ Agent picks up where it left off

Итог: Multi-session feature delivered without repeated context dumping.

Подводные камни

Task graph drifts from reality after manual commits — Run a reconciliation step occasionally — diff completed tasks against actual repo state

Сочетать с: github

Research mode: explore options before committing code

👤 Engineers comparing approaches ⏱ ~30 min intermediate

Когда использовать: You have 2-3 implementation options and want a structured exploration, not yolo.

Поток

Enter research mode

Shrimp research_mode: compare three state management options for my React app — Zustand, Redux Toolkit, Jotai. Note tradeoffs.✓ Скопировано

→ Research notes per option
Convert winner into tasks

I'll go with Zustand. Convert the research conclusion into a task plan.✓ Скопировано

→ Task graph aligned with decision

Итог: Informed decision tracked alongside the tasks to execute it.

Сочетать с: idea-reality

Enforce project rules on every agent turn via Shrimp

👤 Teams with strict coding standards ⏱ ~15 min beginner

Когда использовать: Agents keep forgetting your ESLint config or architecture constraints.

Поток

Set project rules

In Shrimp, set rules: (1) no any in TypeScript (2) all network code in the services/ layer (3) tests next to source.✓ Скопировано

→ Rules saved
Execute with rules in-context

Execute next task; Shrimp injects the rules automatically.✓ Скопировано

→ Agent respects rules without reminders

Итог: Consistent output across a long session.

Комбинации

Сочетайте с другими MCP — эффект x10

shrimp-task-manager + github

Each completed task becomes a commit with a linked PR

After each execute_task, commit with Conventional Commits format; when all tasks in the plan done, open a draft PR.✓ Скопировано

shrimp-task-manager + overture

Visual plan in Overture, persistent execution in Shrimp

Use Overture to approve the plan graph; then hand tasks to Shrimp for tracked execution.✓ Скопировано

Инструменты

Что предоставляет этот MCP

Инструмент	Входные данные	Когда вызывать	Стоимость
plan_task	description: str, rules?: str[]	Start of a feature	LLM tokens
execute_task	task_id?: str (defaults to next open)	Drive next step	LLM tokens + tool calls
research_mode	question: str	Pre-code exploration	LLM tokens
continuous_mode		When you trust the plan and want hands-off	can be high
set_rules	rules: str[]	Project onboarding	0

Стоимость и лимиты

Во что обходится

Квота API: Uses your LLM's quota
Токенов на вызов: Plan: 2-5k. Execute: depends on task
Деньги: Free
Совет: Avoid continuous_mode until you've validated a plan step-by-step for the current project

Безопасность

Права, секреты, радиус поражения

Хранение учётных данных: None intrinsic; wherever the executed tasks reach

Исходящий трафик: Task state stored locally in your configured data dir

Устранение неполадок

Частые ошибки и исправления

Agent forgets task graph after restart

Ensure the data dir path in .mcp.json is stable and not /tmp

Проверить: cat the state file — tasks should be present

Continuous mode goes rogue

Kill the session; Shrimp keeps state so you can inspect and resume selectively

Build step fails on npm install

Requires Node 18+. Shrimp has native deps on some setups — check platform notes

Альтернативы

mcp-shrimp-task-manager в сравнении

Альтернатива	Когда использовать	Компромисс
sequentialthinking-tools	You want in-session planning, not persistent tasks	No cross-session memory
Linear MCP	You want tasks in your team's issue tracker	Cloud/paid; team overhead

Ещё

Ресурсы

📖 Читать официальный README на GitHub

🐙 Открытые задачи

🔍 Все 400+ MCP-серверов и Skills