Claudia is a task management system for AI agents. Through CLI and MCP server interfaces, it enables AI assistants such as Claude to autonomously manage complex work.
Purpose
When AI agents execute complex, multi-step tasks, they need to track progress, coordinate multiple agents, remember context across sessions, verify acceptance criteria, and organize work into sprint and project hierarchies. Claudia provides structured task tracking for these needs.
Main Capabilities
- Hierarchical tasks: supports parent-child task structures
- Dependencies: dependencies can be defined between tasks, and ready tasks can be identified
- Sprint management: organize work into sprints
- Acceptance criteria verification: acceptance criteria can be verified one by one before a task is completed
- Multi-agent coordination: supports multiple AI agents working on the same project simultaneously, with an optimistic locking mechanism
- Context storage: up to 64KB of JSON storage per task, used for cross-session memory
- MCP server: integrates with tools such as Claude Code
- Dashboard: provides a kanban board (with draggable columns) and sprint card view, can be filtered by project, and can archive completed tasks
Usage
- Requires the Bun runtime (v1.0+).
- A standalone binary can be built.
- Start the MCP server for Claude Code: add configuration in .mcp.json in the project root directory, or in development mode run directly without building. After adding the configuration, restart Claude Code to connect.
- After starting the dashboard, open it in the browser at http://localhost:3333 to view the kanban and sprint views.
- Agents can use MCP tools for task operations. Recommended pattern: task_ready (list tasks whose dependencies are satisfied), task_claim (atomically reserve a task), task_transition (move to in_progress), task_context_merge (save progress), task_transition (move to completed); on failure, use task_release to let other agents retry.
Task States
Valid state 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) If a task has no acceptance criteria defined, verification can be skipped.
MCP Tool Reference
- Create and start a task (single operation, automatically claimed for the current agent)
- Complete a task (optional summary)
- Get the current agent's workspace context (including claimed tasks)
- Transfer a task to another agent
- Release a task back to pending and explain the reason
- List all tasks whose dependencies are not satisfied
- List all tasks whose dependencies are satisfied and can be started
- Overwrite task context (maximum 64KB)
- Deep merge into existing context
- Mark an acceptance criterion as verified
- Get a task's verification progress
- List all projects
Applicable Scenarios
- Progress tracking for complex multi-step tasks
- Multiple agents collaborating on the same project
- AI workflows that need cross-session memory context
- Tasks that require acceptance criteria verification
- Scenarios that need to organize work into sprint and project hierarchies
Limitations
- Requires the Bun runtime (v1.0+)
- Task context is limited to 64KB
- License is GPLv3
- Official website:https://claudiacli.com/