jobber-mcp

Jobber Client Hub via fpx (no MCP)

This reads the customer side of Jobber: the hub a business shares with you, at clienthub.getjobber.com. It is not the Jobber Developer API — that one serves the business running on Jobber and needs an OAuth app you cannot register as their customer. See references/why-not-the-api.md.

clienthub.getjobber.com sits behind a Cloudflare managed challenge that fingerprints the TLS client, so plain curl and Node get 403 Just a moment even with a current Chrome User-Agent and the full browser header set. fpx issues the request from inside your own signed-in tab, which has already cleared the challenge. There is no server-side path; the bridge is not optional here.

One-time setup

npm install -g @fetchproxy/cli               # provides `fpx`
fpx profile add jobber --domain getjobber.com # fetch capability only — no cookie scope needed

The first fetch prints a pair code to stderr; approve it in the Transporter extension popup. Pairing persists — every later call reuses it.

Requirements: the Transporter extension installed in Chrome, an open clienthub.getjobber.com tab signed into the hub, and the extension's Chrome Site access allowing getjobber.com.

Only the fetch capability is declared, deliberately. Cookies ride the tab automatically, so no cookie scope is needed — and widening scope after the first approval leaves fetches working on the old grant while the new capability errors. Everything this skill does is covered by the grant above.

Your hub URL is a credential

Each business gives you a different hub, identified by a UUID:

https://clienthub.getjobber.com/client_hubs//

Anyone holding that URL can read the hub, so treat it like a password: keep it in an env var, never in a committed file or a shell history you share.

Get it from any email that vendor sent you — the "View Details" / "View Invoice" button — or from the address bar of an open hub tab.

export JOBBER_HUB='https://clienthub.getjobber.com/client_hubs/'

One export per business. If two vendors both use Jobber, they are two hubs with nothing in common; there is no combined view and no account that spans them.

Core call

Fetch the page, pipe it through the parser, and you have JSON for jq:

PARSE="$(dirname "$0")/references/parse-clienthub.mjs"   # or an absolute path

fpx get -p jobber "$JOBBER_HUB/appointments" \
  | node "$PARSE" appointments \
  | jq '.'

The parser is dependency-free — a bare node runs it, no install step.

jq alone cannot do this job: the hub is server-rendered HTML, and its two page families store data two different ways (JSON islands for appointments, plain cards for everything else). The parser hides that split behind one interface.

Ready-to-run recipes — next visit, unpaid invoices, totals, a single record — are in references/recipes.md.

Pass the right kind — the failure is silent otherwise

appointments reads embedded JSON; the other three read HTML cards. Point the appointments reader at the invoice page and it finds the page's one island — an unrelated referral widget — which parses cleanly and contains no invoices. It looks like "you have no invoices" rather than like a bug.

The parser warns on stderr whenever it returns an empty list, for exactly this reason. An empty result with no warning means the page genuinely had none.

Exit codes

The parser follows the fpx convention, so a pipeline can branch on either:

From fpx itself: 2 bridge down, 3 bot wall, 4 upstream non-2xx.

fpx get -p jobber "$JOBBER_HUB/invoices" > page.html || {
  echo "fpx failed ($?) — is Chrome running with a signed-in hub tab?" >&2; exit 1; }
node "$PARSE" invoices < page.html

What this cannot do

Read-only, by design and by capability:

• No writes. Submitting a work request, approving a quote or confirming an appointment are form POSTs with CSRF and, on some flows, a Turnstile token read from the DOM. fpx has no DOM-read verb, so it cannot complete them.
• No PDF or file downloads. Invoice and quote PDFs are served as Content-Disposition: attachment; the bridge does fetch(), not navigation, so these URLs can only be resolved for you to open, never fetched.
• No payments. Paying an invoice means entering card or bank details. Never automate that — open the hub and do it yourself.