ClickUp MCP: The Free Server Has a Daily Call Ceiling

ClickUp, the project management app, now runs its own MCP server. MCP is the standard that lets an AI assistant call another product’s tools directly, so Claude or Cursor can read and change your tasks without you pasting anything.

The setup is genuinely easy. You paste one URL, https://mcp.clickup.com/mcp, into your client. You approve an OAuth screen. That is the whole install, and it works on every ClickUp plan including the free one.

Then you ask your agent to tidy up next sprint, and somewhere in the middle of the answer it stops.

That is not a bug. ClickUp meters this server by the day, not by the minute, and the meter covers your entire workspace. On the free plan the allowance is small enough that one long agent session can finish it. You also cannot check how much you have left, because ClickUp does not show you.

The official server, and where it lives

ClickUp publishes the server at https://mcp.clickup.com/mcp. It is remote and hosted, so there is no npm package, no local process and no config file full of secrets.

ClickUp’s developer documentation still carries a public beta notice on it. The help center article does not mention beta at all, which is the first of two places these two documents disagree. The second one matters more and it is further down.

It works with any client that speaks the protocol properly. ClickUp spells out the bar for a home-grown one, which is JSON-RPC 2.0 over HTTP, OAuth 2.1 with PKCE, and the MCP specification. If you are new to all of this, our plain explanation of MCP covers the protocol without the acronyms.

No admin has to switch anything on inside ClickUp first. You connect, you approve, the server tells your client which tools exist.

What it actually does

The published tool list is organized into eight groups. Search, task management, time tracking, workspace hierarchy, members and assignees, chat, docs, and time in status reporting.

The coverage inside those groups is real work, not a demo:

• Tasks: create, read, update and delete, plus bulk create and bulk update, custom fields, subtasks, dependencies, task links, tags, comments and threaded replies, file attachments, and moving a task between lists.
• Docs: create a doc, list its pages, read page content, add a page, edit a page.
• Time: start and stop timers, log entries after the fact, pull entries for one task or many.
• Hierarchy: read the whole space, folder and list tree, create and update lists and folders.
• Reporting: how long a task sat in a status, and the same figure across a whole list.

That last one is quietly the best tool in the set. “How long do tasks usually spend in QA review” is the kind of question people never ask, because pulling it by hand is tedious enough that nobody bothers.

One thing to notice, and it is different from Trello. The ClickUp server can delete a task. Atlassian’s official Trello server has no destructive operations at all, a difference we covered in the Trello MCP write-up. Here, an agent that misreads you can remove work permanently.

The number nobody quotes

Here is the part every existing guide leaves out.

Without the Everything AI add-on, ClickUp does not limit MCP requests per minute. It limits them per rolling 24 hours, and the ceiling depends on your plan.

Three details make that ceiling bite harder than it reads:

1. It applies to the entire workspace. ClickUp’s help center says every MCP client connected to the same workspace draws from the same pool. Your Claude session and your colleague’s Cursor session spend the same allowance.
2. Every request counts. Reading a task counts. Listing a folder counts. It is not limited to anything AI-flavored.
3. The window starts at your first call, and it cannot be reset. Spend the lot by 11am and you are locked out until the same hour tomorrow, returning 429 RATE_LIMIT_EXCEEDED until then.

Now put that next to how agents behave. One instruction like “find everything overdue in the product space and move it to next sprint” is not one call. It is a search, a hierarchy read, then an update for each task. A tidy-up across thirty tasks is comfortably thirty-something calls.

And you are flying blind while it happens. Asked whether you can monitor how many calls the workspace has used, ClickUp’s own FAQ answers in three words. Not at this time.

Buying the Everything AI add-on removes the daily bucket and puts MCP on the same per-minute limits as the public API, which are far more generous. That is a real fix. It is also a paid add-on to solve a problem the free tier creates.

ClickUp’s two documents disagree about the free limit

This is worth stating plainly, because it decides whether the free tier is usable.

Checked on August 28, 2026, ClickUp’s help center article lists the daily allowances as 100 calls on Free Forever, 300 on Unlimited, 1,000 on Business, 2,500 on Business Plus and 5,000 on Enterprise.

The developer documentation, last updated August 13, 2026, lists two numbers instead. 50 calls per 24 hours on Free Forever, and 300 per 24 hours on Unlimited and above.

They cannot both be right. The free plan is either 50 or 100, and the paid plans are either capped at 300 across the board or scale up to 5,000.

What to assume

Assume the smaller number until your own workspace proves otherwise. Both documents agree the limits are subject to change, and ClickUp says as much in a note on the developer page. Plan for 50 and be pleasantly surprised.

The honest reading is that this is a beta being tuned in public. That is fine. It is just not something to build a team habit on without checking first.

What it cannot touch

The published tool list has no entry for any of these:

• Goals and key results: if your quarter runs on OKRs inside ClickUp, the agent cannot see them.
• Views: no creating, updating or deleting views, and no fetching tasks scoped to one.
• Task templates: no browsing them and no creating a task from one.
• Webhooks: nothing to wire up automation from the agent side.
• Checklists: subtasks are covered, the checklist inside a task is not.
• Spaces: you can read the hierarchy and create lists and folders, but there is no create or update tool for a space itself.

There is one more limit that catches people out. If you have connected external apps to ClickUp for Connected Search, the MCP server cannot search them. ClickUp is explicit that Connected Search only works from inside the workspace.

None of this makes the server bad. It makes it a task, doc and time server rather than a whole-workspace one, and that is a reasonable thing for a beta to be.

The community server is not free any more

Since it appeared at the end of 2025, the answer to “the official one is missing something” has been taazkareem/clickup-mcp-server. It covered goals, views, templates, webhooks, checklists, sprints and full workspace CRUD, and it was open source.

That changed. The current repository says the project moved from open source to a paid model, and the npm package now publishes under a proprietary license. Every setup snippet in its readme carries a license key, either as an X-License-Key header or a CLICKUP_MCP_LICENSE_KEY environment variable.

Read from the project’s own checkout links on August 28, 2026, the license is $9 a month or $59 a year, each covering three device activations. A one-time lifetime license is sold alongside them, listed at $89 behind a discount code that expires in days, so check what it costs on the day you look.

That is not a complaint. Maintaining an integration this wide is real work and the pricing is modest. But it changes the shape of the decision, and none of the guides currently ranking for this term mention it at all. The free fallback people assume is there has a paywall on it now.

Still free, and smaller

If you need self-hosting without a license, three MIT-licensed servers are alive at the time of writing. Nazruden/clickup-mcp-server and arslan8122/clickup-mcp were both last touched in May 2026, and Chykalophia/ClickUp-MCP-Server---Enhanced in August 2026. All three are far smaller in scope, so check the tool list against your actual need before committing. A fourth, nsxdavid/clickup-mcp-server, has not been pushed to since April 2025.

The auth trade is the real decision

The official server accepts OAuth and nothing else. ClickUp’s FAQ is blunt about it, and its reasoning is sound. Keys are static, broadly scoped and hard to revoke.

That is right for a person using Claude at a desk. It is wrong for a few real cases:

• A service account running scheduled work with no human at the OAuth screen.
• A shared team server where you want one credential rather than every member connecting their own.
• Anything on a machine where opening a browser to approve a consent screen is awkward.

A self-hosted server solves those, because it talks to ClickUp’s public API with a key you control. It also moves you onto a different meter, since ClickUp states that MCP limits and public API limits are two separate things, the API being per minute and per token. If the daily ceiling is your problem, that is the actual reason to self-host.

Verify that against your own workspace before you spend anything on it. And whichever way you go, an agent with write access to a task board deserves a moment’s thought, which is what our guide to vetting an MCP server is for.

Which one you want

For most readers it is the official one. The maintenance cost of a hosted server is zero, and the two things people usually discover they need, goals and views, are worth checking for before you assume you need them.

Copy this prompt

Connect to ClickUp and answer these three, then stop.
1. How many tasks are open in each space?
2. Which open tasks are past their due date?
3. How long do tasks usually sit in each status on my main list?
Tell me roughly how many tool calls that took.

Run that on day one. It exercises search, hierarchy and the time-in-status tools in one go, and the last line gives you a rough feel for how fast the daily allowance goes.

What neither of them reaches

Both servers stop at the edge of ClickUp. That is the correct scope for them and it leaves a real gap.

Your board holds what you and your team decided to do. It does not hold the article that changed your mind, the thread where someone explained the approach you copied, or the post you saved three months ago and half remember now.

That material is scattered across the places you read, and the agent reading your tasks cannot see any of it. So it plans well and reasons from nothing.

Where ContextBolt fits

ContextBolt sits on the other side of that gap.

ClickUp holds what you planned. ContextBolt holds what you saved, the web pages you kept and the posts you saved on X, Reddit and LinkedIn, captured as you save them. It tags and clusters them as they arrive, and Pro exposes an MCP endpoint so Claude can search that library in the same session it reads your tasks. The two servers sit side by side in one agent, one for the plan and one for the reading behind it.

The design contrast with this post is worth naming. ContextBolt’s bookmark server ships six tools, search_bookmarks, list_clusters, get_cluster_bookmarks, get_recent_bookmarks, export_bookmarks and save_bookmark. A server you leave connected in every session has to be cheap to leave connected. And the endpoint is personal to your account, so there is no workspace pool for a colleague’s session to drain.

Pro is $6 a month, month to month, and includes the MCP endpoint, which searches the library by meaning, plus cloud sync and no cap on bookmarks. Basic is free and capped at 150 bookmarks, with AI tagging and clustering included, which is enough to find out whether you would use it.

Worth doing?

Yes, and it takes about two minutes because there is nothing to install.

Go in knowing what kind of thing it is. The official ClickUp MCP server is a capable set of hands on a short leash. Use it for the work an agent is genuinely good at, which is triage, restructuring, drafting docs and answering questions about your own board.

Do not wire it into anything that runs unattended on the free plan. A scheduled job that reads your workspace every hour will eat the day’s allowance before lunch, and you will not get a warning, because there is no meter to warn you.

And check the numbers yourself before you plan around them. Two ClickUp pages currently give different limits for the same free plan, which tells you this is still moving.

Then give the agent the other half. Install ContextBolt, free up to 150 bookmarks, and the same Claude session that moves your tasks can pull up the article that explains why they are there.