Jobber MCP Server: Ask Your Jobber Account Questions in Plain English
A free connector with 12 read-only tools. It lets Claude, ChatGPT or Microsoft Copilot read your live Jobber data. Which invoices are past due. What you billed last quarter. What you did for this customer before. It runs on your machine. It cannot change anything.
A scripted demo with made-up data. The real one answers from your account.
You:|
Harbour Coatings$8,42061 days
Delta Roofing$3,15044 days
Kestrel Property$1,89033 days
$13,460 past 30 days, across 3 clients
Tools used: overdue_invoices
July$71,20038 invoices
August$64,88031 invoices
September$83,41045 invoices
$219,490 collected across the quarter
Tools used: revenue_summary
Jobs14since 2024
Quotes3 openoldest 22 days
Invoices1 unpaid$8,420
Last visit 12 August. Last payment 4 July.
Tools used: find_client + client_history
Install it yourself, or have us do it
Install it yourself
Free, MIT, and about fifteen minutes if you are comfortable in a terminal. You register a Jobber developer app, paste one block into your AI client, and sign in once.
We register the app, wire it to your account, and walk one person on your team through it. Then we show you what it still cannot answer, which is usually the useful part.
Three steps
1. Register a Jobber developer app
At developer.getjobber.com, signed in as a Jobber admin. Copy the client ID and secret. Set the redirect URI the README gives you.
2. Add it to your AI client
One config block. The README has the exact text for each client. Claude Desktop, claude.ai, Claude Code, Cursor, ChatGPT Developer mode and Microsoft Copilot Studio.
3. Sign in once
Run the authenticate tool. Your browser opens, a Jobber admin approves it, and the tokens are encrypted on your machine. Then ask it something.
All 12 tools
Taken from the connector's own source. Eight read your account, four manage the connection.
find_clientread
Search clients by name, email, or phone. Returns contact info and addresses.
client_historyread
A client's jobs, quotes, invoices, and payments.
overdue_invoicesread
Unpaid invoices past due date, oldest first, total owed at the top.
quotes_outstandingread
Sent quotes awaiting a client response, with quote age and amounts.
jobs_summaryread
Jobs grouped by status over a date range, with counts and totals.
revenue_summaryread
Paid invoices grouped by month and quarter, with totals.
schedule_lookupread
Visits and assessments for a date range, grouped by day.
requests_inboxread
New and unscheduled requests, as two separately paged sections.
authenticateconnection
Start the Jobber sign-in. A Jobber admin approves it once.
auth_statusconnection
Check whether it is connected, and when the token expires.
logoutconnection
Revoke at Jobber where possible, then clear the local tokens.
get_audit_logconnection
Read the connector's own log of every call it has made.
Nothing matches. Try a broader word.
The meter most Jobber integrations ignore
Jobber meters its API by query cost, not just by request count. You get 10,000 points. They refill at 500 a second. A second ceiling caps you at 2,500 requests every 5 minutes. We measured one dashboard load at 13,456 to 20,762 points, across 41 to 52 calls. One screen can spend twice the whole budget.
Then there is the part that catches people. When Jobber throttles you it answers with HTTP 200 and puts the error in the body. Retry code reads that as success and hands back nothing.
What this connector does about it
Every tool declares its top cost and checks the budget before it spends.
It reads the throttle signal inside a 200, waits for the refill, and retries once.
Page sizes are capped, so one question cannot drain the bucket.
If the wait would run long it says so, instead of hanging.
The connector reads what Jobber's API exposes. Plenty of the questions owners care about most sit outside that. Crew utilisation. Hours by person. Cost and profit per job, which needs your accounting data joined to Jobber. A dashboard that stays fast, which needs its own database syncing in the background.
That is the work we do for money. It is a separate conversation, and you only have it if you want to. The free connector stays free.
There is a paid hosted product, Zapier's MCP, two meta-connectors, and more than a dozen repos on GitHub. We compared every one of them, including this one, and said which fits what.
Anonymized here. Ask on the call and Dan will tell you whose shop each one is.
"They just got how our business works."
"We needed our Jobber account connected to our internal system, and Danilo's team understood the workflow almost immediately. They handled the integration, cleaned up a few things we were struggling with, and made the whole process much easier for our team."
HVAC Company OwnerJobber CRM Integration
"We stopped doing a ton of this stuff manually."
"Our biggest problem was that our office team was constantly moving information between systems. Danilo's team built the integration for us and automated a big part of the workflow. It's one of those projects where you wonder why we didn't do it sooner."
Home Services Business OwnerCustom CRM & Workflow Automation
Jobber MCP server FAQ
We have not seen one in Jobber's developer materials as of August 2026. Jobber's own AI work is inside the app: the receptionist, voice, and campaign features. None of it connects your account to Claude or ChatGPT. Every way to do that today comes from a third party, this connector included. Check developer.getjobber.com before you build, because this moves fast.
The connector is free and MIT licensed. You run it on your own machine, on your own Jobber app, and nothing routes through us. The code in this repository stays MIT, and new releases of it stay MIT. We make our money building the systems it cannot build. That is a separate conversation, and only if you want it.
No. Version 1 ships read tools only. It can look, count, and summarize. It cannot create a job, send an invoice, or move a visit. The worst failure is a wrong answer, not a wrong action. Write tools will come one at a time, each behind an approval step, and each logged. Walk away from any tool that asks for your Jobber password instead of an admin-approved sign-in.
Jobber gates full API access to its top plan, so check your plan before you start. Marketplace apps and Zapier work lower down, and they cover trigger-and-action automation. Answering plain-English questions from live account data runs through the API. Jobber changes plan pricing often, so read their pricing page rather than any blog post, this one included.
Yes, with one requirement. Claude can run the connector locally on your machine. ChatGPT connects through Developer mode. Microsoft Copilot Studio connects through Streamable HTTP. Both of those need the connector reachable over the internet, not just on your laptop. It ships both modes, so all three work.
Jobber meters its API by query cost, not just by request count. You get 10,000 points, refilling at 500 points a second, plus a separate ceiling of 2,500 requests every 5 minutes. One dashboard load we measured cost 13,456 to 20,762 points across 41 to 52 calls. Jobber also signals throttling inside an HTTP 200 response, where standard retry code reads it as success. This connector tracks the budget, waits out short refills, and stops before it drains the bucket.
On your machine. You register your own Jobber developer app. A Jobber admin approves it. The tokens are encrypted at rest under your home directory. Nothing is held on our servers, because the connector runs on yours. Every call it makes is written to a local log you can read.
Start with the free connector. Hire us if you hit its edges.
Takes about a minute, and you get the Jobber developer app steps either way.