Taskade

Implements the Model Context Protocol (MCP) to establish bidirectional communication with the Taskade platform, enabling real-time synchronization of task state, updates, and changes. Uses MCP's resource and tool abstractions to expose Taskade's task management API as standardized endpoints that LLM clients can invoke without custom integration code. Maintains persistent connections to stream task mutations and project state changes to connected AI agents.

Exposes Taskade's create, read, update, delete operations as MCP tool definitions with JSON schema validation. Each operation (create task, update project, delete workflow) is registered as a callable tool with input schemas that validate parameters before transmission to Taskade. Implements error handling and response mapping to translate Taskade API responses back into structured tool outputs that LLMs can parse and reason about.

Implements MCP resource endpoints that expose Taskade workspaces, projects, and task hierarchies as queryable resources. Uses MCP's resource URI scheme to allow clients to request specific projects, task lists, or workspace metadata by identifier. Resources are returned as JSON documents that include full task trees, project settings, team members, and workflow definitions, enabling agents to build rich context before executing task operations.

Provides MCP integration points for Taskade's native AI agents and workflows, allowing agents to invoke other agents, chain operations, and manage workflow state through MCP tool calls. Agents can spawn subtasks, delegate work to other agents, and coordinate multi-step processes by invoking workflow execution tools that Taskade exposes via MCP. Supports passing context and task state between agent invocations.

Implements MCP's streaming capability to push real-time task mutations, status changes, and workspace events to connected clients. Uses server-sent events or WebSocket-like patterns over MCP to notify agents of task updates without polling. Agents can subscribe to specific projects, task lists, or event types and react to changes as they occur, enabling reactive workflows that respond to external task modifications.

Enables agents to query and link tasks across multiple Taskade workspaces and projects through MCP resource requests. Agents can retrieve task references, create cross-project dependencies, and synchronize state across workspace boundaries. Implements namespace resolution to disambiguate task IDs across workspaces and provides tools to create and manage inter-workspace task relationships.

Exposes Taskade's custom field definitions and metadata schemas through MCP resources, allowing agents to discover and interact with workspace-specific task properties. Agents can read and write custom fields (e.g., priority levels, cost estimates, custom enums) by querying the workspace schema and using CRUD tools that understand custom field types. Supports dynamic schema discovery so agents can adapt to different workspace configurations without hardcoding field names.

Implements query parameter support in MCP resource requests to filter tasks by status, assignee, priority, date range, and custom fields. Agents can construct resource URIs with query parameters (e.g., taskade://tasks?status=open&assignee=user123) to retrieve filtered task lists without fetching and filtering client-side. Supports complex queries combining multiple filters and sorting options.

Exposes task dependencies and relationships (blocks, is_blocked_by, relates_to, duplicates) as MCP resources. Agents can query task dependencies to understand task prerequisites and blocking relationships. Implements dependency graph as queryable resource, enabling agents to understand task sequencing and identify critical path items.

Implements MCP subscription mechanism for real-time task events (task created, updated, completed, commented, assigned). Agents can subscribe to task events and receive notifications as they occur, enabling reactive agent behavior. Uses MCP's subscription/notification pattern to push events to connected agents without polling.