ClickUp

ClickUp's official MCP shipped in 2026 after a beta period that started late 2025, and the surface area has grown from a handful of tools at launch to roughly 49 across 14 categories. That makes it the largest first-party project-management MCP available right now. You're getting search, task management, bulk operations, attachments, comments, tags, relationships, time tracking, workspace hierarchy, members, chat, docs, and time-in-status reporting. Install is a single command: `claude mcp add clickup --transport http https://mcp.clickup.com/mcp`. Cursor and other Streamable HTTP clients connect the same way. OAuth handles auth, no API token to copy around. The server runs remote so there's no Docker or Node package to maintain locally. The bulk operations bit is what makes this MCP genuinely useful versus just being a smaller surface for the ClickUp API. Asking an agent to "find all overdue tasks assigned to me across these three projects, push their due dates out by a week, and add a comment explaining why" is a one-shot operation. The same workflow in the ClickUp UI is fifteen clicks and a mental note not to miss any. For teams running large workspaces, the time savings on triage alone justify the integration. Time tracking is the other category worth calling out. The MCP exposes time-tracking start, stop, and time-in-status queries. If your team's billing or sprint planning relies on those, surfacing them through Claude means you can do retroactive time entry conversationally. ("I worked on the auth refactor from 2 to 4pm yesterday, log it against the Backend project.") The known issues from the public feedback thread are around task creation with assignees and custom fields, plus OAuth persistence in automated agent contexts. If you're scripting unattended workflows, validate the assignee writes are going through. For interactive agent use, these aren't blockers. For ClickUp teams, this is a default-install. For teams evaluating Asana vs ClickUp vs Linear, ClickUp's MCP currently has the deepest tool inventory of the three, which matters if you want agents doing real work in the tool rather than just reading from it.