Claudia

io.github.yuvalsuede由 GitHub 帳號 yuvalsuede 發布v0.1.3更新於 Oct 25, 2026

Task management for AI agents with hierarchical tasks, dependencies, sprints, and coordination.

已驗證STDIO僅桌面效率與工作流開發者工具

概覽

AI 產生的概覽

Claudia 是本地任務管理伺服器，讓 AI 助理建立、追蹤、協調並驗證階層式任務、相依性、衝刺與專案。

可以做什麼
建立、讀取、更新、刪除與批次編輯任務，支援階層樹狀結構以及待處理、進行中、驗證、已完成之間的狀態轉換，並為每個任務保存最多 64KB 的 JSON 內容與附證據的驗收條件驗證。管理相依性與協調：新增或移除阻塞關係，列出被封鎖與可執行的任務，並為多個代理原子性地認領、釋放、移交或放棄任務。執行複合操作，例如一步建立並開始任務、附摘要完成任務，或取得目前代理的工作區。將工作組織成衝刺與彼此隔離的專案，包括選擇目前專案並依工作目錄自動辨識。
適用情境
適合助理處理長時間、多步驟的工作且需要跨工作階段保留任務狀態時，也適合多個代理共用同一專案、需要避免重複認領時。也適合希望透過命令列與網頁看板管理含衝刺與驗收檢查之任務的團隊。
執行需求
以本機程序透過 stdio 執行，使用 bunx 從 npm 套件 claudia-cli 啟動；README 要求 Bun 執行環境 v1.0 以上。使用本機資料庫檔案，預設位於使用者家目錄下，可用環境變數 CLAUDIA_DB 變更。未宣告需要帳號、API 金鑰或請求標頭。
安裝前請注意
此伺服器會寫入本機資料庫：任務、衝刺與專案都可被建立、更新、切換狀態與刪除，刪除具破壞性。任務內容每個任務最多保存 64KB JSON，請勿在其中存放機密資訊。README 的快速開始會從原始碼複製並建置，並透過指令碼安裝 Bun，執行前請先審閱。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Claudia，將其新增到工作區。
  2. 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

Claudia

[License: GPLv3] [Bun] [MCP]

A task management system built for AI agents.

Official website: https://claudiacli.com/

Claudia provides structured task tracking with a Model Context Protocol (MCP) server, enabling AI assistants like Claude to manage their own work through hierarchical tasks, dependencies, sprints, and acceptance criteria verification.

[Claudia Dashboard Demo]

Why Claudia?

AI agents need a way to:

  • Track progress across complex, multi-step tasks
  • Coordinate when multiple agents work on the same project
  • Remember context between sessions (64KB JSON storage per task)
  • Verify work against acceptance criteria before completion
  • Organize work into sprints and project hierarchies

Claudia provides all of this through both a CLI and MCP server interface.

Quick Start

bash
# Install Bun if you haven't alreadycurl -fsSL https://bun.sh/install | bash
# Clone and buildgit clone https://github.com/yuvalsuede/claudia.gitcd claudiabun installbun run build
# Initialize and start using./claudia db init./claudia task create --title "My first task"./claudia task list

Try the Demo

See Claudia in action with sample data:

bash
# Seed demo project with sample tasks and sprintsbun run seed:demo
# Open the web dashboard./claudia @@ --port 3333

Then open http://localhost:3333 in your browser to explore the kanban board and sprint views.

Screenshots

Task Board

[Claudia Tasks View]

Sprint Management

[Claudia Sprints View]

Features

FeatureDescription
Hierarchical TasksParent-child relationships with tree visualization
State MachineValidated transitions: pending → in_progress → verification → completed
DependenciesBlock tasks until prerequisites complete, with cycle detection
SprintsGroup tasks into time-boxed work periods
Multi-ProjectIsolated task namespaces with auto-detection from working directory
Agent Memory64KB JSON context storage per task
Acceptance CriteriaDefine and verify requirements before task completion
Multi-Agent CoordinationAtomic task claiming, optimistic locking, conflict detection
Web DashboardVisual kanban board with project/sprint filtering
MCP ServerDrop-in integration with Claude Code and other MCP clients

Installation

Prerequisites

  • Bun runtime (v1.0+)

From Source

bash
git clone https://github.com/yuvalsuede/claudia.gitcd claudiabun install

Build Standalone Binary

bash
bun run build# Creates ./claudia binary
# Optional: install globallycp claudia ~/.bun/bin/

Usage

CLI Commands

Task Management
bash
# Create a taskclaudia task create --title "Implement feature X" --priority p1
# Create with acceptance criteriaclaudia task create --title "Add login" --acceptance-criteria "Has email field" --acceptance-criteria "Has password field"
# List tasksclaudia task listclaudia task list --status in_progress --priority p0,p1
# Show task detailsclaudia task show <task-id>
# Update a taskclaudia task update <task-id> --title "New title" --priority p0
# Transition statusclaudia task transition <task-id> --to in_progress
# Delete a taskclaudia task delete <task-id> --force
Task Hierarchy
bash
# Create a subtaskclaudia task create --title "Subtask" --parent <parent-id>
# View task treeclaudia task tree              # Full treeclaudia task tree <task-id>    # Subtree from task
Task Context (Agent Memory)
bash
# Set context (overwrites)claudia task context-set <task-id> '{"key": "value"}'
# Merge context (deep merge)claudia task context-merge <task-id> '{"additional": "data"}'
# Get contextclaudia task context-get <task-id>
Dependencies
bash
# Add dependency (task depends on blocker)claudia task depends <task-id> --on <blocker-id>
# Remove dependencyclaudia task undepends <task-id> --on <blocker-id>
# Show dependenciesclaudia task deps <task-id>
# List blocked tasksclaudia task blocked
# List ready tasks (all deps satisfied)claudia task ready
Sprints
bash
# Create a sprintclaudia sprint create --name "Sprint 1" --start 2024-01-15 --end 2024-01-29
# List sprintsclaudia sprint list
# Show sprint with tasksclaudia sprint show <sprint-id>
# Activate a sprintclaudia sprint activate <sprint-id>
Projects
bash
# Create a projectclaudia project create --name "My Project" --path /path/to/project
# List projectsclaudia project list
# Select active projectclaudia project select <project-id>
# Show current projectclaudia project current

Web Dashboard

bash
# Open dashboard in browserclaudia @@
# Custom portclaudia @@ --port 8080

The dashboard provides:

  • Tasks View: Kanban board with drag-and-drop columns
  • Sprints View: Sprint cards with progress indicators
  • Project Filter: Scope views to specific projects
  • Clear Completed: Archive finished tasks

MCP Server Integration

Start the MCP server for use with Claude Code:

bash
claudia mcp
Claude Code Configuration

Add to .mcp.json in your project root:

json
{  "mcpServers": {    "claudia": {      "command": "/path/to/claudia",      "args": ["mcp"]    }  }}

Or for development (without building):

json
{  "mcpServers": {    "claudia": {      "command": "bun",      "args": ["run", "/path/to/claudia/src/mcp/server.ts"],      "cwd": "/path/to/claudia"    }  }}

After adding the config, restart Claude Code to connect.

Available MCP Tools

Compound Operations (Recommended for agents)

ToolDescription
task_startCreate and start a task in one operation
task_finishComplete a task with optional summary
task_workspaceGet current agent's workspace context
task_handoffTransfer task to another agent
task_abandonRelease task back to pending

Task Management

ToolDescription
task_createCreate a new task
task_readGet task by ID
task_updateUpdate task fields
task_deleteDelete a task
task_listQuery tasks with filters
task_transitionChange task status
task_treeGet hierarchical task view

Coordination

ToolDescription
task_claimAtomically claim a task
task_releaseRelease a claimed task
task_blockedList blocked tasks
task_readyList ready tasks
task_dependency_addAdd task dependency

Verification

ToolDescription
task_verifyMark criterion as verified
task_verification_statusGet verification progress

See the full MCP tools reference below for complete documentation.

Task Workflow

┌─────────┐     ┌─────────────┐     ┌──────────────┐     ┌───────────┐│ pending │────▶│ in_progress │────▶│ verification │────▶│ completed │└─────────┘     └─────────────┘     └──────────────┘     └───────────┘     │                │                    │                    │     ▼                ▼                    ▼                    ▼┌─────────┐                         ┌──────────┐│ blocked │                         │ archived │ (terminal)└─────────┘                         └──────────┘

Valid transitions:

  • pending → in_progress, blocked, archived
  • in_progress → pending, verification, completed, blocked, archived
  • verification → in_progress, completed, blocked, archived
  • blocked → pending, in_progress, archived
  • completed → in_progress, archived
  • archived → (terminal state)

Tasks can skip verification if no acceptance criteria are defined.

Multi-Agent Coordination

Claudia supports multiple AI agents working concurrently on the same project.

Task Claiming

typescript
// Claim before workingconst result = await task_claim({ task_id: "uuid", agent_id: "agent-1" });if (result.success) {  // Task is yours - proceed} else {  // Already claimed by another agent}
// Release when done or on failureawait task_release({ task_id: "uuid", agent_id: "agent-1" });

Optimistic Locking

typescript
// Read task firstconst task = await task_read({ id: "uuid" });
// Update with version checkawait task_update({  id: task.id,  title: "Updated",  version: task.version  // Fails if modified by another agent});

Recommended Pattern

  1. task_ready - List tasks with satisfied dependencies
  2. task_claim - Atomically reserve a task
  3. task_transition - Move to in_progress
  4. task_context_merge - Save progress
  5. task_transition - Move to completed
  6. On failure: task_release - Let another agent retry

Configuration

Environment VariableDescriptionDefault
CLAUDIA_DBDatabase file path~/.claudia/tasks.db

Development

bash
# Development modebun run dev
# Run testsbun test
# Type checkingbun run typecheck

Exit Codes

CodeMeaning
0Success
1General error
2Resource not found
3Conflict (version mismatch)
4Validation error
5Storage error

MCP Tools Reference

Compound Operations
task_start

Create and start a task in one operation. Auto-claims for the current agent.

json
{  "title": "Task title",  "description": "Optional description",  "priority": "p0|p1|p2|p3",  "parent_id": "optional-parent-uuid",  "acceptance_criteria": ["criterion 1", "criterion 2"]}
task_finish

Complete a task with optional summary.

json
{  "id": "task-uuid",  "summary": "Optional completion summary"}
task_workspace

Get current agent's workspace context including claimed tasks.

json
{  "include_completed": false}
task_handoff

Transfer task to another agent.

json
{  "task_id": "task-uuid",  "to_agent_id": "target-agent",  "notes": "Optional handoff notes"}
task_abandon

Release task back to pending with reason.

json
{  "task_id": "task-uuid",  "reason": "Why abandoning"}
Task CRUD
task_create
json
{  "title": "Required title",  "description": "Optional",  "status": "pending",  "priority": "p0|p1|p2|p3",  "parent_id": "uuid",  "sprint_id": "uuid",  "tags": ["tag1", "tag2"],  "assignee": "name",  "acceptance_criteria": ["criterion"]}
task_read
json
{ "id": "task-uuid" }
task_update
json
{  "id": "task-uuid",  "title": "New title",  "version": 1}
task_delete
json
{ "id": "task-uuid" }
task_list
json
{  "status": ["pending", "in_progress"],  "priority": ["p0", "p1"],  "parent_id": "uuid",  "sprint_id": "uuid",  "assignee": "name",  "limit": 100,  "offset": 0}
task_transition
json
{  "id": "task-uuid",  "to": "in_progress"}
task_tree
json
{  "id": "optional-root-uuid",  "depth": 5}
Bulk Operations
task_create_many
json
{  "tasks": [    { "title": "Task 1" },    { "title": "Task 2" }  ],  "parent_id": "optional-common-parent",  "sprint_id": "optional-common-sprint"}
task_update_many
json
{  "ids": ["uuid1", "uuid2"],  "updates": {    "priority": "p1",    "assignee": "agent-1"  }}
task_transition_many
json
{  "ids": ["uuid1", "uuid2"],  "to": "completed",  "skip_invalid": true}
Dependencies & Coordination
task_dependency_add
json
{  "task_id": "blocked-task",  "depends_on_id": "blocking-task"}
task_dependency_remove
json
{  "task_id": "task-uuid",  "depends_on_id": "dependency-uuid"}
task_dependencies
json
{ "task_id": "task-uuid" }
task_blocked

List all tasks with unsatisfied dependencies.

task_ready

List all tasks ready to work on (dependencies satisfied).

task_claim
json
{  "task_id": "task-uuid",  "agent_id": "claiming-agent"}
task_release
json
{  "task_id": "task-uuid",  "agent_id": "releasing-agent"}
Context Storage
task_context_set

Overwrite task context (max 64KB).

json
{  "id": "task-uuid",  "context": { "any": "json data" }}
task_context_merge

Deep merge into existing context.

json
{  "id": "task-uuid",  "context": { "additional": "data" }}
task_context_get
json
{ "id": "task-uuid" }
Verification
task_verify

Mark an acceptance criterion as verified.

json
{  "task_id": "task-uuid",  "criterion_id": "criterion-uuid",  "evidence": "Optional verification evidence"}
task_verification_status

Get verification progress for a task.

json
{ "task_id": "task-uuid" }
Sprints
sprint_create
json
{  "name": "Sprint 1",  "start_at": "2024-01-15",  "end_at": "2024-01-29"}
sprint_list
json
{ "include_archived": false }
sprint_show
json
{ "id": "sprint-uuid" }
sprint_update
json
{  "id": "sprint-uuid",  "name": "New name",  "status": "active"}
sprint_delete
json
{ "id": "sprint-uuid" }
sprint_activate
json
{ "id": "sprint-uuid" }
Projects
project_create
json
{  "name": "Project name",  "description": "Optional",  "path": "/optional/directory/path"}
project_list

List all projects.

project_read
json
{ "id": "project-uuid" }
project_update
json
{  "id": "project-uuid",  "name": "New name"}
project_delete
json
{ "id": "project-uuid" }
project_select
json
{ "id": "project-uuid" }
project_current
json
{ "cwd": "/optional/path/for/autodetect" }

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

License

GPLv3 - see LICENSE for details.

來源：README.md，提交 8898fa8

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包，因此其工具未知。

版本歷史

1
  1. v0.1.3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