The Fieldwork MCP connector lets an AI assistant read your Fieldwork data directly and answer questions about it in plain English — customers, work orders, routes, agreements, invoices, and the reports built into Fieldwork. It works with any AI assistant that supports MCP (Model Context Protocol), not just one specific product.
You don't need to learn any commands or know which report holds an answer. Just ask your question the way you'd ask a coworker who has Fieldwork open, and the assistant finds the right data.
Getting started
Ask a direct question and the assistant will look up the answer for you. For example:
"Can you find the customer account for [company or contact name]?"
"How many missed work orders do we have this month?"
"What were our cancellations last week?"
You can search by a full name, a partial name, an account number, or a phone number — the assistant will match on whatever you give it and confirm the record before pulling details.
What you can ask for
Look up a record A customer, service location, work order, invoice, or agreement. Give a name or number and the assistant finds it, then can pull full detail — billing info, service locations, recent work orders and invoices — as a follow-up.
Get a count or list Cancellations in a given period, missed appointments, incomplete work orders, units where a tenant refused service, overdue invoices, and similar.
Check a number Total accounts receivable, this month's completion rate, top-performing sales rep, a specific customer's balance.
See a breakdown Any of the above, split by technician, route, service type, branch, or time period.
How it works
There are two ways the assistant answers a question:
Direct lookups — for one specific record (a customer, a work order, an invoice), it fetches that record and related details directly.
Reports — for anything aggregate (totals, trends, breakdowns), it runs one of the reports built into Fieldwork. There are dozens available, covering receivables and collections, technician production, sales and commissions, service agreements, compliance tracking, and profitability. A handful of these are pre-built summary dashboards that return ready-made KPIs instead of raw rows.
You never need to name a specific report — describe what you want to know, and the assistant maps it to the right one.
Going further
Once you're comfortable with basic lookups, the connector can handle more involved requests:
Combining reports — a question like "which customers look like they might churn" pulls agreements and receivables data together and looks for accounts with both a lapsed renewal and a significant overdue balance, a pattern that isn't visible in either report alone.
Longer time spans — most reports cover one month or one year at a time. Ask for a longer range (a full year of month-by-month comparisons, for instance) and the assistant runs the report repeatedly and totals the results, rather than refusing.
Re-cutting the same data — ask to see an existing answer broken down a different way (by service type instead of technician, for example) without repeating the whole analysis.
Open-ended analysis — a broad question like "where are we losing money operationally" pulls together whichever reports are relevant (financials, receivables aging, job-level profitability, and so on) and comes back with the findings ranked by size rather than a dump of raw data.
Shareable output — ask for a chart, summary document, or spreadsheet instead of a chat answer when you want something to hand off to someone else.
Tips for better results
Say which period you mean. "This month," "last week," and specific date ranges all work — being specific avoids a wrong guess on an ambiguous request.
Ask for a breakdown explicitly. Any total can usually be split by technician, branch, service type, or customer — just say which.
Follow up. Treat the first answer as a starting point. "Who's overdue" naturally leads to "which of those are worth prioritizing" — keep going rather than starting over.
Flag anything that looks off. If a number doesn't match what you expected, say so — the assistant will re-check rather than assume the first answer was right.
Things to keep in mind
Report time windows are bounded. Most reports run one month or one year at a time. A longer request gets answered by running the report multiple times and adding it up, which takes a little longer than a single-period question.
Large results are summarized, not dumped in full. If a request matches thousands of records, the assistant will read through them and surface totals and the rows relevant to your question, rather than listing everything. Ask for the complete list if that's what you actually need.
Test or sample data can look like real data. On an account that's still being set up, placeholder or test records can appear in results alongside real ones. Worth a second look on anything high-stakes until your account is fully live.
Examples of what you can ask
You want to know... | Ask something like... |
A specific customer's details | "Pull up [customer]'s account" |
What's overdue and by how much | "Give me a summary of our AR" |
Who to follow up with on payment | "Who are my most overdue customers?" |
Jobs that didn't happen | "How many missed work orders this month?" |
Contracts that ended | "What were our cancellations last month?" |
Whether techs are arriving on time | "Run the quality of service summary" |
Whether jobs run long or short | "Compare scheduled time vs. actual time worked" |
Who's selling the most | "Who's our top sales rep based on agreements?" |
Where revenue is leaking | "Where are we losing money operationally?" |
Accounts at risk of leaving | "Which customers look like they might churn?" |
A shareable version of any answer | "Turn this into a report / chart / spreadsheet" |