Karea MCP

What starecz/karea-mcp does

starecz/karea-mcp exposes Karea work-management data through MCP. It is organized into nine grouped tools covering read operations, task writes, notes, project changes, meetings, resources, assistant requests, deletion, and action help. Together, the groups provide 70 actions for managing development work and the information attached to it.

Tasks are the central workflow. Agents can find and inspect tasks, create or edit them, apply the same change to multiple tasks, close work, record completed work, or create an in-progress task. Tasks can include categories, priorities, deadlines, descriptions, tags, notes, closing requisites, subtasks, Jira issue keys, resources, and AI coding-session references.

The server also handles supporting project information. It can create and share projects, manage categories, maintain sticky notes, store task markdown, preserve titled AI context entries across sessions, track open questions, manage resources and meetings, schedule reminders, and produce an activity recap. The assistant action accepts a natural-language request that may read or modify Karea tasks.

How it works

Install and run the package as an MCP process. The client invokes one of the grouped tools with an action and a params object; for example, task actions are sent through the task-writing group rather than registered as separate MCP tools by default. karea_help returns the full JSON schema and description for a named action, or lists the available actions when no action is supplied.

Read operations generally require a useful filter. Task, question, meeting, resource, and sticky-note list actions reject unfiltered requests and return the accepted filters so an agent can retry. Project listing and pending-reminder checks are exceptions. Actions that identify tasks accept visual IDs, names, or UUIDs where supported.

Setup and configuration

Run the package with:

npx -y karea-mcp

Set KAREA_API_KEY to an API key from Karea's API-key settings. KAREA_URL can point the server at the Karea service URL; the Claude Code example sets it to https://karea.app.

For Claude Code, register the server in ~/.claude.json or use claude mcp add. For Cursor, place the MCP definition in .cursor/mcp.json within a project or in the global ~/.cursor/mcp.json. Both configurations launch npx with the -y flag and package name karea-mcp.

The optional KAREA_MCP_LEGACY_TOOLS=1 setting restores registration of the individual actions as separate tools instead of the grouped tool layout.

Tools and capabilities

• Read projects, tasks, subtasks, notes, sticky notes, task markdown, AI context, questions, resources, meetings, reminders, Jira links, coding sessions, and recaps.
• Create, edit, batch-update, close, delete, or quickly log tasks and subtasks.
• Add closing requisites and toggle them before completing a task.
• Write human-readable task notes, long-form markdown, and cross-session AI context.
• Create, update, upload, delete, and attach text or file resources.
• Link tasks to Jira issues, meetings, questions, and AI coding sessions.
• Create meetings, attach tasks or questions, and manage reminders.
• Ask Karea's assistant to act on task information using natural language.

Destructive operations can permanently remove projects, categories, tasks, notes, resources, meetings, and other records. Several of these actions require an explicit confirmation parameter.

Limitations and notes

A single unfiltered list request is not accepted for most collections; provide at least one supported filter. The server's default interface groups actions, so callers must use the correct action name and parameter object. Task documents and AI context serve different purposes: markdown is the task's long-form document, while context stores titled AI working-memory entries covering the task's history.

The karea_mark_reminder_done action completes both the reminder and its underlying task. Deleting a category also deletes its tasks, and deleting a project removes its contained tasks, categories, notes, and history.