io.github.chrischall/jobber-mcp

Jobber Client Hub MCP

A Model Context Protocol server that connects Claude to the Jobber Client Hub — the customer portal that home-service businesses (pest control, lawn care, HVAC, cleaning) use to send you appointments, quotes and invoices.

[!WARNING] AI-developed project. This codebase was built and is actively maintained by Claude Code. No human has audited the implementation. Review all code and tool permissions before use.

This is the customer side, not the business side

Jobber has two surfaces, and they share nothing:

If you run a business on Jobber, you want the Developer API instead — jobber-mcp by justinvogel covers that surface. This server is for being someone's customer, and the reasoning is written up in skills/jobber-fpx/references/why-not-the-api.md.

What you can do

• "When is the exterminator coming next?"
• "Do I owe Queen Bee's anything?"
• "Show me every invoice they've sent this year."
• "What did I ask them to do in my last work request?"

Read-only, and not by omission — see Why there are no writes.

Requirements

• Claude Desktop or Claude Code
• Node.js 22 or later
• The ContextMint Bridge browser extension, installed from its releases page (Chrome: load the chrome zip unpacked. Safari isn't available yet — it will ship inside the ContextMint app, which has no public download — so use Chrome for now), with Site access allowing getjobber.com, and a signed-in Client Hub tab. ContextMint Bridge is the fetchproxy browser extension under its new name, from the same maintainer — fetchproxy's own README points to it. Its source is public at https://github.com/nullnet-app/contextmint-bridge: build it yourself, or check a release zip against the .sha256 file published beside it (shasum -a 256 -c contextmint-bridge-chrome-<version>.zip.sha256).
• A Client Hub link from a provider — the "View Details" button in any of their emails

Setup

{
  "mcpServers": {
    "jobber": {
      "command": "npx",
      "args": ["-y", "@chrischall/jobber-mcp"],
      "env": { "JOBBER_HUB_ID": "<the UUID from your hub URL>" }
    }
  }
}

Your hub URL looks like clienthub.getjobber.com/client_hubs/<UUID>/. The UUID is the whole credential — anyone holding it can read the hub — so treat it like a password.

Several providers, several hubs (there is no combined view):

"env": {
  "JOBBER_HUBS": "[{\"label\":\"queenbee\",\"hubId\":\"…\"},{\"label\":\"greenworx\",\"hubId\":\"…\"}]"
}

Then pass hub: "greenworx" to any tool. With one hub configured you never need the argument.

Installed from the .mcpb bundle in Claude Desktop, the same three settings appear as Client Hub ID, Client Hub label and Several Client Hubs (JSON) — paste the JSON array above into the last one.

Tools

jobber_list_* keeps each record's metadata rows raw in details rather than forcing a schema. Which rows a card shows depends on its state — an unpaid invoice carries a balance row a paid one does not — so a fixed schema would invent fields for some records and drop rows from others.

Why the browser bridge is not optional

Verified live 2026-08-09: clienthub.getjobber.com sits behind a Cloudflare managed challenge that fingerprints the TLS client, not the User-Agent. curl and Node both get 403 with the Just a moment interstitial, and keep getting it when handed a current Chrome UA and the full browser Accept* header set. The identical request from inside a real tab returns 200.

There is also no JSON API to fall back on: the hub is a server-rendered Rails app that makes zero API calls to its own origin. clienthub.getjobber.com/api/graphql exists and answers introspection, but it serves the same staff schema as the Developer API — it is not a client-facing endpoint.

Hosting on mcp-host

mcp-host can run this server as a bridged child, which is how it reaches claude.ai and phones. The hub pages are still fetched by your own signed-in browser tab: the ContextMint Bridge extension pairs with the hosted child over the network, so the Cloudflare challenge sees your real browser, never the datacenter. Nothing replays a lifted cookie, which is why the @fetchproxy/bootstrap "lift the session once" pattern is deliberately not used here.

mint.yaml carries what a registration needs: register it with bridge: true and give JOBBER_WS_PORT as the bridgePortEnv (the runner injects a per-registration concentrator port through it; locally it defaults to the fleet-shared 37149 via readPortEnv), keep state.dataDir so the bridge identity and pairing survive restarts, and leave egress empty, since the process itself never dials getjobber.com.

Why there are no writes

The hub can submit work requests, approve quotes and pay invoices. None of them are here:

• Those flows are form POSTs carrying CSRF tokens and, on some paths, a Turnstile token read from the DOM. The bridge does fetch(), not DOM reads, so it cannot complete them — a write tool would fail unpredictably rather than work.
• Paying an invoice means handling card or bank details. That belongs in your hands, in your browser, not in an agent's tool call.

Without the MCP

skills/jobber-fpx/ does the same reads from a shell with the fpx CLI — no server process. Same bridge, same pages, one command.

Development

npm install
npm run build
npm test

The suite mocks the network entirely; tests/server-boot.test.ts additionally boots the real built artifacts — including the bundle in a directory with no node_modules, as the .mcpb runs — and drives a full initialize + tools/list handshake.

Acknowledgement of terms

1. This server accesses your own Client Hub. Every request is dispatched through your own signed-in browser session via the ContextMint Bridge extension, reusing the session you already have. It does not — and cannot — reach anyone else's hub.

2. Jobber's Terms of Service govern your use of this server, exactly as they govern your direct use of the hub in a browser. Review them, and stop using this server if your use of it would not comply.

License

MIT