What the agent does
This is for a bookkeeping or accounting firm that closes the books for dozens of small-business clients each month. Each client has a List in a ClickUp Space called "Monthly close", and each month the firm creates the same tasks in it from its checklist: bank and card reconciliations, accrued expenses, payroll entries, a review, and sending the client their reports. The partner wants to know, a few working days into the month, which closes are behind and why, without opening every List. The run names one connection, clickup, ClickUp's remote MCP server signed in as the firm's operations manager.
The agent starts with ClickUp's clickup_get_workspace_hierarchy tool to find the "Monthly close" Space and the List for each client. For each List it calls clickup_filter_tasks for the tasks due in the month being closed, and clickup_get_list to read the List's statuses, so it knows which status means done in that List. For each open task past its due date it reads the comments with clickup_get_task_comments, because that is where a bookkeeper writes "waiting on October bank statement from client" or "payroll report not received".
It sorts each open task into one of three reasons, using the comments: waiting on the client (a document, an answer, an approval), waiting inside the firm (the review step, a teammate), or no reason given. Then it ranks the clients: closes with the most overdue tasks first, and within a client, the task that blocks the most others, such as the bank reconciliation that the review depends on. It does not judge whether an entry is right; it reads status, dates and comments.
Last, it writes the report into a new ClickUp Doc with clickup_create_document, named for the month, with one section per client that is behind: the tasks overdue, their assignees, days late, the reason and the latest comment quoted with its author and date. Clients whose close is on track get one line each at the end. It creates no task, changes no status and posts no comment.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Use the ClickUp connection. Find the Space 'Monthly close'; each List in it is one client. For each List, read its statuses with clickup_get_list and its tasks due between 2026-10-01 and 2026-10-31 with clickup_filter_tasks. For every task not in a closed status and past its due date, read its comments. Classify each as waiting_on_client, waiting_internal or no_reason, quoting the comment you used. Create one ClickUp Doc named 'October 2026 close status' with a section per client that has overdue tasks, worst first: task, assignee, due date, days late, reason, latest comment with author and date. End with one line per client that is on track. Do not create or update tasks, change statuses or post comments. Return the Doc's link and the count of clients behind.",
"connections": [
"clickup"
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
The run creates one Doc in the firm's ClickUp workspace and returns its link and the number of clients behind in result. The partner reads it at the start of the week and decides who calls which client; the bookkeepers see their own late tasks with the reason the agent took from their comments.
What the partner checks before acting on it: a task listed as overdue that was finished but never moved to the closed status, which the bookkeeper fixes; a reason the agent read wrongly from a vague comment, which is quoted so it is quick to spot; and clients with no List at all for the month, which means the checklist was never created. Every List, task and comment the agent read is in the session's transcript.
Run it on every close week
In the console, a Project automation with a schedule trigger runs a saved agent that names clickup and holds the firm's rules: the Space name, which statuses count as done, the three reasons. A cron of 0 9 3-10 * * in the firm's timezone runs it each morning from the 3rd to the 10th, when most closes should finish; the month to report on goes in the automation's prompt.
Each morning's run makes a new Doc, so the partner can open the 4th's and the 7th's side by side and see which closes moved. A firm that wants one Doc can tell the agent to add a dated page to the month's Doc with clickup_create_document_page instead.
The use case this job follows, with its automation and trigger: Produce a report or file, no repo.
Why run it on Rush?
The close checklist already lives in ClickUp, and the useful part, why a task is late, lives in free-text comments. A ClickUp dashboard counts overdue tasks; it cannot tell "waiting on the client's bank statement" from "I forgot". The agent reads the comments the way the partner would, and quotes them, so its reasons can be checked in seconds.
The ClickUp sign-in stays in Rush's connection, signed in with OAuth, rather than a personal API token in a script on someone's laptop; ClickUp's MCP server takes OAuth only. Every run has its own computer and a record of each List and comment it read and the Doc it made, and the session lists what its model calls cost.
Frequently asked questions
Will the agent chase clients for missing documents?
Not in this run. It lists what each close is waiting on. A firm that wants drafts of the reminder emails can add a second step that writes them into the Doc for a bookkeeper to send.
Can it change task statuses that are wrong?
ClickUp's MCP server has clickup_update_task, but the prompt forbids it. A task that is done but still open is listed, and the bookkeeper who owns it closes it.
What if each client's checklist is slightly different?
That is fine. The agent reads whatever tasks are in each List for the month and reads each List's own statuses, so a client with an extra sales-tax step is reported on that step too.
Does it see client files attached to tasks?
It reads task fields and comments. It does not open attachments unless the prompt asks it to, and this one does not, so bank statements and payroll files stay unread.