Jobber MCP Server Comparison: Every Option for Claude, ChatGPT and Copilot (2026)

Why your choice of Jobber MCP option matters

An MCP (Model Context Protocol) server is a translator between an AI assistant and a software system's API. Point one at Jobber's GraphQL API and Claude or ChatGPT can answer questions like "which invoices are past 30 days?", "summarize our history with this customer," or "which quotes went out this month and never converted?" from your live account, instead of you clicking through nine screens.

Three things separate the options, and none of them show up in a feature list screenshot:

• Who holds your credentials. A hosted product or meta-connector holds your Jobber OAuth token on its servers. A self-hosted server keeps it on your machine. For a business whose customer list and invoice history is the business, this is not a detail.
• Whether it can write. A read-only connector can answer questions but cannot change a job, invoice, or schedule; the worst failure is a wrong answer. A write-capable one with no approval step can reschedule a crew or send an invoice on a model's misunderstanding.
• Whether it survives Jobber's rate limiting. This is the one almost everyone ships wrong, and it gets its own section below.

One plan-level gate applies to every option on this page: custom API access requires Jobber's top plan (Plus). Jobber changes pricing regularly, so check Jobber's pricing page rather than any number a blog post quotes.

Every current option, side by side

Verified against each product's public pages and repositories on August 20, 2026.

1. JobberToClaude: hosted, polished, closed

jobbertoclaude.com: $47/month after a 7-day trial, verified August 20, 2026.

What it does. A private hosted MCP server per customer, claiming 85+ actions across Jobber, with OAuth-based connection. You sign up, connect your Jobber account, and add their endpoint to Claude. It is the fastest route from zero to "Claude answers questions about my Jobber."

What to weigh. It is closed source, run by a team that does not name itself beyond "we run our own field service business," and your Jobber OAuth token lives on their infrastructure. None of that makes it bad; it makes it uninspectable. You cannot verify what the write actions can do, how failures are handled, or what happens to your data in transit. At $47/month it also costs more per year than most shops' entire software-tinkering budget.

Who it fits. A non-technical owner who wants this working today, accepts the subscription, and is comfortable with a third party holding the keys.

2. Zapier MCP: honest, shallow

zapier.com/mcp/jobber exposes Zapier's existing Jobber catalog through MCP: 11 triggers, 6 write actions, 2 searches.

What it does. Trigger-and-action automation, now callable by an AI assistant. "Create a client," "add a job", the same building blocks Zapier always had.

What to weigh. It is not natural-language querying of your account. "Show me all quotes over $5,000 that have not converted in 30 days" is not a Zapier action; it is a query with filtering and aggregation, which is exactly what a purpose-built MCP server does and a trigger catalog does not. Zapier also bills each MCP tool call against your task quota, and agent conversations are chatty by nature. And your Jobber connection lives on Zapier's side.

Who it fits. A shop already paying for Zapier that wants its existing automations reachable from an AI chat. For that job it is the most trustworthy name on this page.

3. Pipedream and viaSocket: the meta-connectors

Pipedream and viaSocket both list Jobber among thousands of apps reachable through their generic MCP gateways.

What they do. Wrap their prebuilt workflow actions for Jobber in an MCP interface. Setup is quick and the free tiers are real.

What to weigh. Same shape as Zapier with less field-service depth: generic wrappers over the raw API, no domain model of visits versus jobs versus requests versus quotes, and connector quality that varies by app. Pipedream was acquired by Workday in late 2025 and its focus is visibly shifting enterprise. Credentials, again, live on their side.

4. The open-source field on GitHub

Fourteen public Jobber MCP repos existed on GitHub as of August 20, 2026, the oldest from February 2026. We read all of them. The honest summary:

• Ten of the fourteen went quiet within days of their first commit. Two are archived, one was deleted, and the most-starred repo in the field has four stars and belongs to an account that has published over a hundred near-identical auto-generated "MCP servers" for different platforms. Its install instructions reference a Jobber settings page that does not exist.
• Tool coverage ranges from 5 to 18 tools, mixing reads and writes. Several ship write actions (create clients, create jobs) with no confirmation step. One exposes a raw GraphQL escape hatch, which makes both access and API costs unpredictable.
• Authentication is the sharpest divide. Roughly half ask you to paste an access token that expires within the hour, guaranteeing the thing breaks mid-conversation. A handful implement the real OAuth flow with token refresh. One repo, otherwise the most security-conscious of the field, has an OAuth refresh token accidentally committed into its public git history, which is the category's maturity level in one sentence.
• Exactly one of the fourteen handles Jobber's throttling signal correctly. More on why that matters below, because it is the difference between a demo and a tool.

Who it fits. A technical owner or a shop with a developer on call, who can read the repo before trusting it, and who treats "maintained" as a requirement rather than a nice-to-have. Free and self-hosted is genuinely the right architecture; the current implementations are just not finished.

The DIY variant of this option is building your own tools in n8n and exposing them over its MCP trigger. Real, flexible, and a build project rather than an install.

Does Jobber have an official MCP server?

No. As of August 2026, Jobber's developer center has no MCP pages, its app marketplace lists no MCP integrations, and there has been no announcement. Jobber's AI investment is in-product: AI Receptionist, voice, and marketing features inside Jobber itself, useful, but none of it connects your account to the assistant you already use. The demand is visible in Jobber's own community forum, where owners have been asking for and hand-building exactly this since June.

When a platform vendor eventually ships first-party MCP, it tends to cover the vendor's preferred surface, on the vendor's terms. The value of an open connector (your own tools, your own permissions, writes behind your own approval step) survives that day. We watched the same sequence play out in legal software, where the vendor's "official MCP" turned out to cover research content, not your own account data.

The rate-limit problem almost everyone ships wrong

This is the section to read if you only read one, because it is invisible in every feature table and decides whether the thing works on a Tuesday afternoon.

Jobber meters its GraphQL API by query cost: a budget of 10,000 points per app-and-account pair, refilling at 500 points per second, on top of a plain request ceiling. Every response reports what the query cost and what is left. Two facts make this brutal for AI use:

• Report-shaped questions are expensive. One production dashboard load our team measured cost 13,456 to 20,762 points against the 10,000-point budget. An AI conversation asking "give me this month's numbers by crew" has the same shape. A connector with no per-tool cost caps drains the whole budget on one question and then fails everything else.
• Jobber signals throttling inside an HTTP 200. The response says success; the error is in the body. Standard retry libraries read it as success and return garbage. Of the fourteen open-source repos, exactly one checks for it.

The full measurements are in our team's Jobber API rate limits breakdown. The design consequences (per-tool cost budgets, page caps, refill-aware backoff, and why a raw query tool is a mistake) are in the Jobber MCP server deep dive. That deep dive is the spec for the open-source connector we are building; when it ships, it will be the first in this category with cost budgeting built in, and this comparison will be updated to include it with the same scrutiny applied to everyone else.

Decision framework

Whichever you pick: start read-only, connect a test window before a busy Monday, and never paste a long-lived credential into a chat window.

All prices, tool counts, and repository states verified August 20, 2026 against public pages, listings, and GitHub. These change. If you maintain one of the options covered here and a detail is incorrect, email office@oktopeak.com and we will correct it.