What the agent does
This is for a recruiting agency that places accountants, financial analysts and controllers with mid-sized companies. Each client search is a group on a monday.com board called Candidates, and each candidate is an item with a Stage status, a Recruiter person column, the Role and a Last contact date. A candidate who hears nothing for a week takes another offer, and a client who hears nothing about a submitted candidate assumes the search has stalled. The managing partner wants one list on Monday morning: every candidate in an active stage that nobody has touched in seven days, by recruiter, with a short check-in drafted for each.
The agency connects monday.com once with POST /me/connections {"type": "mcp", "name": "monday", "url": "https://mcp.monday.com/mcp"}, signed in as a member who can see every search; monday.com's docs say admins approve MCP connections beyond the third-party agents approved in advance. A run reads the board's columns, groups and owners with get_board_info, pages through every item with get_board_items_page, and for each candidate in Submitted, First interview, Final interview or Offer reads the item's updates with get_updates. The later of the Last contact date and the newest update is that candidate's last touch.
A candidate whose last touch is more than seven days old goes on the list under their recruiter, with the client search, the role, the stage, the days since the last touch, and one line from the newest update so the recruiter remembers where it stood. For each one it drafts a check-in under 80 words: to the candidate for Submitted and interview stages, asking whether anything has changed on their side, and for Offer a note the recruiter can send the client asking for a decision date. Drafts use the candidate's first name and the role's title, nothing else from their file.
It posts no update, changes no column and moves no item: create_update and change_item_column_values are monday.com tools that write, and the prompt forbids them by name with every other tool that writes. monday.com's docs say every MCP tool call runs as a GraphQL API request and counts toward the account's daily API call limit, so the prompt reads each page of the board once and fetches updates only for candidates in an active stage.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Find candidates on our monday.com board Candidates with no update in 7 days, as of 2026-11-16. Use the monday connection. Call get_board_info for columns, groups and owners, then get_board_items_page for every item. For items whose Stage is Submitted, First interview, Final interview or Offer, call get_updates for that item. Last touch = the later of the Last contact date column and the newest update. List every such candidate whose last touch is before 2026-11-09, grouped by the Recruiter column: candidate name, group (the client search), Role, Stage, days since last touch, one line quoting the newest update. Draft a check-in under 80 words for each: for Submitted and interview stages, to the candidate asking if anything has changed on their side; for Offer, a note the recruiter sends the client asking for a decision date. Use the first name and role title only. Write stale-candidates.md and stale-candidates.csv. Do not call create_update, change_item_column_values, create_item, create_notification or any other tool that writes.",
"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
The session returns stale-candidates.md, grouped by recruiter with each draft under its candidate, and stale-candidates.csv for the partner's weekly numbers. The answer says how many candidates are on the list and which recruiter has the most.
What each recruiter checks before sending: candidates they spoke to by phone without logging it, who look quiet but are not; every draft for an Offer, where the client relationship matters more than the reminder; and items sitting in the wrong stage, which the list makes easy to see. The recruiter sends each message from their own email and logs it on the item, which also takes the candidate off next week's list.
Run it on every Monday morning
The partner saves the agent with POST /o/{org}/agents, naming the monday connection, with the stage names, the seven-day rule and the drafting voice as its instructions, and adds a console automation with a schedule trigger (0 7 * * 1, timezone set to the agency's) that runs it. A schedule runs the saved prompt as written, so the saved prompt counts seven days back from today instead of the fixed dates above.
Before a quarterly review with a client, a recruiter can start the same saved agent from the console with the prompt narrowed to that client's group, to see every candidate in that search who has gone quiet.
The use case this job follows, with its automation and trigger: Draft outreach from your CRM.
Why run it on Rush?
The monday.com sign-in stays in Rush's connection, approved once by the agency's admin, instead of a personal API token in a script on one recruiter's laptop. The transcript records every board page and update the agent read, so the partner can see why a candidate was or was not on the list.
Each Monday's run has its own computer and a record of its model cost. A board with several hundred candidates takes many monday.com calls, and the session records them one by one, so when the account nears its daily API limit the agency can see how much of it this job used.
Frequently asked questions
Will the agent post the check-ins for us?
No. It drafts each message in the file and the recruiter sends it, because a message to a candidate or a client speaks for the agency. The prompt forbids the monday.com tools that post updates, change columns or create items.
Why does our admin have to approve the connection?
monday.com's docs say admins approve MCP connections and redirect URIs beyond the third-party agents monday.com approves in advance. The admin approves it once; after that, the member who connected it is the account every run reads as.
Does this use up our monday.com API limit?
monday.com counts each MCP tool call as a GraphQL API request against the account's daily limit. The prompt reads each board page once and fetches updates only for candidates in an active stage, which keeps a weekly check small.
Can it read candidates' CVs?
This prompt does not ask it to. It reads the board's columns and updates, and the drafts use only a first name and a role title, so files attached to items stay out of the run.