tasks

Creates new tasks through the Model Context Protocol by exposing a standardized tool interface that LLM clients can invoke. The server implements MCP's tool definition schema, allowing Claude, other LLM agents, or MCP-compatible clients to define task properties (title, description, priority, due date) and persist them to a backend store. Works by registering task creation as a callable tool with JSON schema validation, enabling type-safe task instantiation from natural language or structured prompts.

Exposes tasks as MCP resources that clients can read and subscribe to, implementing the MCP resource protocol for standardized task retrieval. The server maintains a resource URI scheme (e.g., 'task://all' or 'task://filter?status=open') and returns task lists as structured JSON. Clients can request resources directly or subscribe to updates, enabling real-time task synchronization without polling. Uses MCP's resource subscription mechanism to push changes to connected clients.

Modifies existing tasks through MCP tool calls that apply atomic mutations (status changes, priority updates, due date modifications, description edits). The server implements optimistic locking or version-based conflict detection to prevent race conditions when multiple agents or clients update the same task. Updates are validated against the task schema before persistence, ensuring data integrity. Returns the updated task object with new timestamps and version identifiers.

Removes tasks from active view through MCP tool calls that implement soft-delete semantics (marking tasks as deleted rather than purging records). The server maintains a deletion timestamp and optional reason, preserving audit trails and enabling recovery. Hard deletion may be available through separate administrative tools. Deleted tasks are excluded from default list queries but can be retrieved through explicit archived/deleted task resources.

Provides filtered task views through MCP resource URIs with query parameters (status, priority, assignee, due date range, tags). The server parses resource requests and returns filtered task subsets without requiring separate tool calls. Supports common filtering patterns like 'tasks://open', 'tasks://high-priority', 'tasks://due-today'. Filtering is performed server-side, reducing client-side processing and enabling efficient pagination of large task sets.

Exposes the task data model schema through MCP's capability discovery mechanism, allowing clients to understand task properties, required fields, valid enums (for status, priority), and constraints without hardcoding. The server provides a schema resource or tool that returns JSON Schema definitions for task objects. Clients use this to validate inputs before calling task creation/update tools, and to render dynamic UIs based on the schema.

Integrates task management into LLM agent planning loops by exposing tasks as contextual resources that agents read before deciding on actions. The server provides task state snapshots that agents use in chain-of-thought reasoning (e.g., 'I see 3 high-priority tasks, I should focus on the one due today'). Agents can invoke task mutations as part of multi-step plans, with the server tracking task state changes across plan execution. Enables agents to reason about task dependencies and sequencing.