Claudia

Claudia provides a Model Context Protocol (MCP) server that lets AI assistants manage tasks, memory, and coordination through a structured, server-driven interface. You can run the MCP server locally, connect clients like Claude Code, and perform multi-agent task management with memory, dependencies, sprints, and verification built in.

How to use

You use Claudia’s MCP server to enable AI agents to coordinate, track progress, and verify work on complex projects. Start the MCP server from your Claudia installation, then connect your MCP client or Claude Code to the running server. From the client you can create and manage tasks, set memory contexts, define dependencies, organize work into sprints and projects, and verify acceptance criteria before completing tasks.

How to install

Prerequisites: you need Bun installed to run Claudia locally.

# Install Bun if you haven't already
curl -fsSL https://bun.sh/install | bash

# Clone Claudia, install dependencies, and build
git clone https://github.com/yuvalsuede/claudia.git
cd claudia
bun install
bun run build

# Initialize and start using
./claudia db init
./claudia task create --title "My first task"
./claudia task list

Starting the MCP server

To run the MCP server for Claudia, start the server using the built CLI with the mcp command. This launches the MCP server in-process so you can connect MCP clients and Claude Code.

claudia mcp

Configure Claude Code to connect to Claudia MCP

Add an MCP integration entry in Claude Code that points to Claudia as an MCP server and configures how Claude Code should launch or connect to Claudia’s MCP endpoint.

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

Available MCP client examples

If you are developing locally and want to test without building every time, you can reference Claudia’s MCP server via the local command shown above. For development, you can also start the server as a CLI process and maintain connections from your MCP clients.

Web dashboard and demo (optional)**

Claudia provides a web dashboard for visual task management and sprint views when running locally. Seed demo data and open the dashboard in your browser to explore the kanban board and sprint views.

Configuration

The MCP server relies on the Claudia runtime and the client code to manage tasks, memory, and coordination. When you run the MCP server with Claudia, you enable multi-agent coordination, task memory, and verification workflows for your projects.

Troubleshooting

If you encounter connection issues between Claude Code and Claudia MCP, verify that Claudia is running with the mcp command and that Claude Code is configured to connect to the correct local MCP endpoint. Ensure dependencies are installed and the CLI is built for your platform.

Configuration and memory notes

Claudia stores per-task memory as 64KB of JSON context, allowing agents to remember task state between sessions. Use acceptance criteria and task context features to manage verification progress and maintain context across tasks and sprints.

Security considerations

When running the MCP server locally, restrict access to your machine and container environment. Only connect MCP clients that you control, and avoid exposing the server URL to untrusted networks.

Notes

This MCP server is designed to work with Claudia’s task management features, including hierarchical tasks, sprints, dependencies, and multi-agent coordination. Use it together with the CLI and web dashboard to manage complex projects.

Available tools

task_start

Create and start a task in one operation and auto-claim for the current agent.

task_finish

Complete a task with an optional summary.

task_workspace

Get the current agent's workspace context, including claimed tasks.

task_handoff

Transfer a task to another agent with optional notes.

task_abandon

Release a task back to pending with a reason.

task_create

Create a new task with details such as title, description, priority, parent, sprint, and acceptance criteria.

task_read

Get a task by its ID.

task_update

Update task fields like title, priority, or description.

task_delete

Delete a task by its ID.

task_list

Query tasks using filters such as status, priority, and assignee.

task_transition

Change a task's status to in_progress, completed, or other states.

task_tree

Get a hierarchical view of tasks.

task_claim

Atomically claim a task for an agent.

task_release

Release a claimed task back to available.

task_blocked

List tasks with unsatisfied dependencies.

task_ready

List tasks with dependencies satisfied and ready to start.

task_dependency_add

Add a dependency between tasks.

task_verify

Mark an acceptance criterion as verified.

task_verification_status

Get verification progress for a task.

sprint_create

Create a sprint with a name and date range.

sprint_list

List sprints.

sprint_show

Show sprint details including its tasks.

sprint_update

Update sprint properties such as name or status.

sprint_delete

Delete a sprint.

sprint_activate

Activate a sprint.

project_create

Create a new project with a name and path.

project_list

List all projects.

project_read

Read a project by its ID.

project_update

Update a project’s properties.

project_delete

Delete a project.

project_select

Select the active project.

project_current

Show the current project context.