What the agent does
This is for a residential property management company that tracks maintenance on a monday.com board: one item per work order, with the property and unit, a category (plumbing, HVAC, appliance, pest), a priority, the vendor assigned, a status, the date reported, and the vendor's notes posted as updates. Each morning the maintenance coordinator scans the board for orders that have sat too long, which is how a leaking water heater turns into a ceiling repair. The agent does the scan. It connects through monday.com's remote MCP server, once, with POST /me/connections {"type": "mcp", "name": "monday", "url": "https://mcp.monday.com/mcp", "auth": {"mode": "oauth"}}, approved on monday.com's own page; an account admin may have to approve the connection first.
The run reads the board's structure with Get Board Info, so it learns the real column names and status labels instead of assuming them, then every open item with Get Board Items, and the recent updates on each with Get Updates. Get Board Activity tells it when each item last changed status, which matters more than the reported date: an order reported eight days ago that moved to "Vendor scheduled" yesterday is not stalled, and one reported three days ago that has sat in "New" since is.
It compares each open order with the company's response targets, which the prompt states (emergency within a day, urgent within three, routine within seven), and with simpler rules: an order with no vendor a day after it was reported, an order in "Waiting on tenant" for more than five days, an order marked complete by the vendor with no closing note. Each flagged order gets its reason, the last update in a sentence, and what to do next.
For orders stalled with a vendor, it drafts a short follow-up the coordinator can send by text or email, naming the unit, the problem and the date it was reported. It writes the report and posts nothing: no updates on items, no status changes, no notifications. Its tool calls run as GraphQL API requests and count toward the monday.com account's daily API limit, so the prompt reads only open items.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Review the monday.com board 'Maintenance - Work orders' as of 2026-11-13 08:00 America/Chicago. Read its columns and status labels with the Get Board Info tool, then every item not in status 'Closed' with Get Board Items, the updates on each open item with Get Updates, and status changes in the last 14 days with Get Board Activity. Response targets by Priority: Emergency 1 day, Urgent 3 days, Routine 7 days, counted from Date reported to a status of 'Vendor scheduled' or later. Flag: an order past its target; no Vendor one day after Date reported; status 'Waiting on tenant' for more than 5 days; status 'Vendor complete' with no update in the last 3 days. For each flagged order give property, unit, category, priority, vendor, days open, reason, the last update in one sentence, and the next step. For each order stalled with a vendor, draft a follow-up of at most 60 words to that vendor. Write work-orders.md, emergencies first. Create, change and post nothing on the board.",
"connections": [
"monday"
]
}- 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
work-orders.md comes back as an artifact on the session. It opens with the flagged orders grouped by reason, emergencies first, each with its unit, vendor, days open and the next step, then the vendor follow-up drafts, then a one-line count of open orders that are on track. The session's result is a short summary the coordinator can read on a phone before getting to the office.
The coordinator checks each flag against the item before acting, because the board is only as current as the team keeps it: a vendor who came yesterday but has not posted an update looks stalled. They send the follow-ups that still make sense from their own phone or email. Nothing about a tenant's home goes to a vendor without the coordinator choosing to send it, and the report contains no tenant phone numbers unless the board's columns hold them and the prompt asks for them.
Run it on every weekday morning
In the console, a Project automation with a schedule trigger (0 7 * * 1-5 with the company's timezone) can run a saved agent that names the monday connection and holds the response targets, so the report is ready before the coordinator starts. When targets change, editing the saved agent changes every later run.
A company with several regional boards gives each its own automation with the board's name in the prompt, or starts runs from its own scheduler with POST /o/{org}/sessions, one per board, with a webhook URL on the API key that tells each region's coordinator their report is ready.
The use case this job follows, with its automation and trigger: Produce a report or file, no repo.
Why run it on Rush?
monday.com's own automations can notify when a date passes, but stalled maintenance is rarely one column: it is a status that has not moved, an update from the vendor that says "waiting on part" four days ago, and a priority that makes four days too long. The agent reads all of it per item and writes the one line the coordinator needs. The monday.com sign-in stays in Rush's connection, not an API token pasted into a spreadsheet script.
Each run has its own computer and keeps a record of every tool call and its cost, so when a flag is wrong the coordinator can see which update the agent read. The board stays untouched: the instructions forbid writes, and the transcript would show any write the agent attempted.
Frequently asked questions
Our board uses different column names. Does the prompt need to change?
The agent reads the board's columns and labels first with Get Board Info, so close names usually match. Name the columns that carry priority, vendor and dates in the prompt when yours differ much, and the status labels that mean scheduled and closed.
Can it post the follow-up as an update on the item?
monday.com's MCP server has a Create Update tool, so it could, but this agent's instructions keep it read-only. Some companies let it post an internal update tagged for the coordinator; vendors and tenants still hear only from a person.
Will it use up our monday.com API limit?
Each tool call is a GraphQL request that counts toward the account's daily API call limit. Reading open items only, and only 14 days of activity, keeps a daily run small, but a company already near its limit should check its usage in monday.com.
Does it decide what counts as an emergency?
No. It uses the Priority column the team sets and the targets in the prompt. If an item's description sounds urgent but its priority is Routine, the agent can note the mismatch for the coordinator to decide.
More monday.com jobs
- Draft a Marketing Agency's Weekly Client Update From a monday.com Board
- What Happened on My Renovation This Week? A Contractor App Drafts Each Homeowner's Update From the Contractor's monday.com Board
- Which Candidates Have Gone a Week Without an Update? A Recruiting Agency's Monday Check of Its monday.com Board
More jobs like this: Produce a report or file, no repo
- Audit Klaviyo Flows (Klaviyo)
- Follow Up on Overdue QuickBooks Invoices Every Week (QuickBooks)
- Find Where Users Drop Off in a PostHog Funnel (PostHog)