Dart

Exposes Dart task management through the Model Context Protocol (MCP), enabling LLM agents to create, read, update, and delete tasks via standardized MCP resource handlers. Implements bidirectional sync between the MCP server and Dart's backend API, maintaining task state consistency through webhook-based change notifications and polling fallbacks. Uses JSON-RPC 2.0 message framing for stateless request/response handling across stdio, HTTP, or SSE transports.

Enables MCP clients to query and retrieve documents attached to Dart projects, with support for semantic search via embeddings. The server indexes document metadata (title, content snippets, creation date, owner) and can filter by project scope. Implements vector similarity search if Dart's backend supports embedding storage, falling back to keyword/metadata filtering. Returns document objects with content references that LLM agents can use to ground task decisions or generate context-aware documentation.

Exposes Dart's project structure (projects, subprojects, team members, roles) as queryable MCP resources. The server fetches project metadata including name, description, owner, team members, and permission model, enabling agents to understand organizational context before creating or modifying tasks. Implements role-based filtering so agents only see/modify projects they have access to, enforced via Dart's backend permission checks. Returns hierarchical project structures that agents can traverse to find the right project scope for a task.

Allows MCP clients to query the schema of custom fields, task types, and metadata structures defined in a Dart workspace. The server introspects Dart's field definitions (name, type, validation rules, allowed values for enums) and returns them as structured JSON, enabling agents to understand what fields are available before creating or updating tasks. Implements schema caching to reduce API calls and supports dynamic schema updates via webhook notifications when workspace admins modify field definitions.

Enables MCP clients to fetch task comments, activity logs, and conversation threads associated with tasks. The server retrieves comment history with metadata (author, timestamp, edited status) and activity events (status changes, field updates, assignments). Implements threading to group related comments and supports filtering by date range or author. Returns comment objects with author information and timestamps, allowing agents to understand task discussion context and make informed decisions about task updates or escalations.

Provides MCP endpoints for performing multiple task operations (create, update, delete) in a single request, with optional transaction-like semantics to ensure atomicity. The server batches operations and sends them to Dart's backend, rolling back all changes if any operation fails (if supported by Dart's API). Implements request deduplication to prevent duplicate operations if the MCP client retries. Returns results for each operation in the batch, allowing agents to handle partial failures gracefully.

Exposes Dart's AI-native features through MCP, allowing agents to access AI-generated task suggestions, reasoning traces, and context attachments. The server retrieves suggestions generated by Dart's AI engine (e.g., auto-generated subtasks, priority recommendations) and returns them with reasoning metadata (why the suggestion was made, confidence score). Enables agents to inspect and act on AI-generated content, creating a feedback loop where agent decisions inform Dart's AI model. Implements versioning to track suggestion history and allow agents to compare different AI suggestions.

Supports bulk operations (create multiple tasks, update task statuses in batch, link docs to tasks) through MCP tool calls with transactional semantics and per-item error reporting. Implements batching logic that groups operations for efficiency while providing granular error feedback for failed items.