Claude for Small Business Pricing: Which Plan You Need for the Plugin

The Claude for Small Business plugin does not show a separate plugin fee: you need a paid Claude Pro, Max, Ultra, Team or Enterprise plan to install it. Pro is the least expensive published entry point at $20 monthly or $200 for a year. Connected services may have their own subscription or feature charges, so price the full workflow before you commit.

Anthropic's Academy names the eligible plans and shows the installation path. Its marketplace page says the workflows start in approval mode: Claude drafts and stages work before a person sends, posts or pays. That reduces the risk of an unintended action, but it does not make the connector, the underlying business software or a human reviewer's time free.

The plugin price is a plan choice, not a line item

The Small Business plugin is a bundle of skills and connectors for jobs such as an overdue-invoice list, a weekly business brief, month-end preparation and lead follow-up. Anthropic's published instructions say to add it from Customize, then Plugins, then Discover, then Small Business. The instructions require Pro, Max, Ultra, Team or Enterprise. They do not present an extra Small Business plugin subscription on top of the Claude plan, and the plugin can be added in Claude on the web or in the desktop app. The marketplace page describes 43 ready-to-run workflows and 37 partner connectors, among them accounting, payments, CRM, e-signature, design and office-suite tools. Do not infer that the plugin works on a Free account just because some individual connectors are available there.

All those published dollar amounts exclude applicable tax and can change. A sole owner who only wants to test one skill should start by pricing Pro, not by multiplying a Team rate. A business that needs multiple named users, shared control and central billing can price Team. Anthropic says Team requires at least two members, so a one-person business would pay for two Team seats. That changes the minimum monthly benchmark from $20 on Pro to $50 for two standard monthly Team seats, or $480 due up front on an annual Team subscription.

The two routes are not interchangeable. Individual Pro may be inexpensive for a founder experimenting with non-sensitive material, but it does not provide the Team administration model. Team does not make the plugin more accurate or exempt the business from checking outputs. Before choosing, write down who will own each connected account, who can approve actions and whether the business needs a shared workspace. The business-app connection tutorial covers the general connector decision; the question here is what the Small Business plugin itself costs to run.

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

Cost the first workflow end to end

Take an illustrative recruitment agency with one owner and four consultants. The owner wants the plugin's Monday brief to combine pipeline, calendar and cash information. A cautious first month uses one Pro account at $20 rather than buying five seats. The owner connects only the tools required for a briefing that can be checked against the existing records. The published Claude cost is $20 for that month, before tax. Any paid CRM or accounting plan already in use remains a separate bill; a new required feature tier would be a new cost.

Suppose the agency then wants three consultants to run lead follow-up workflows. Four Team standard seats, including the owner, would have a published monthly benchmark of $100, before tax. On annual billing the monthly equivalent is $80, but the year is paid up front at $960. These are illustrative seat sums using Anthropic's public rates, not a quote for the agency. The financial comparison should include the time to review each proposed message and the cost of any connected service plan changes. A $20 plugin experiment has become a controlled four-user workflow, so the buyer must revisit the permissions and approval owner too.

For a second quick sum, imagine a four-fee-earner law firm that gives two people Team standard seats to test a contract summary skill. The published annual benchmark is $480 up front for two seats, before tax. If the firm adds a premium seat for one heavy user later, the full-year rate difference is $80 per month equivalent, with a prorated charge when upgraded midterm. The firm should not buy premium because the plugin says “contract review”; it should first record whether the standard user actually hits a usage limit while processing authorised documents.

Which connected services are included?

The plugin contains instructions and references to connectors, but the underlying business accounts are separate. Anthropic's Academy lists tools used by each skill, including accounting, payments, CRM, email and design services. Some tasks work with several alternatives; others require specific data. A connector may be available without a separate Claude connector charge while the connected service still requires a paid account, a feature tier or usage-based billing. The Academy explicitly warns that generating designs or staging sends may depend on the plan held with the connected tool.

Build a service inventory before pressing Add. Use four columns: skill to test, source account, permission needed and possible non-Claude cost. A filled-in illustrative row could read: “invoice-chase; accounting account plus mail account; read invoices and draft messages; existing software subscriptions, no new tier confirmed.” Another row could read: “canva-creator; design account; create designs from an approved brief; confirm whether the current design plan supports the action.” The words “confirm” are deliberate: do not claim a partner fee is zero, or a feature is included, until the partner's own current plan page or administrator confirms it.

One practical example is a marketing agency with a design subscription and a CRM. It asks the plugin to prepare a campaign calendar and draft designs. The Claude subscription may be the only new subscription, but the workflow could still stop when the design account lacks a required capability. If that happens, record the exact blocked action and the partner's proposed upgrade price. Compare that upgrade with a simpler workflow that prepares copy and a brief for a designer. A plugin skill is useful only if the connected account can perform the relevant step at a price the owner accepts.

The cheapest first test may need no connector at all. Anthropic's marketplace page says that without connectors you can upload a spreadsheet or statement and most workflows still run. An owner who wants to judge the invoice-chase or cash-flow-snapshot skill before touching a paid accounting tier can export last month's aged-debtors report, upload it, and compare Claude's ranked list with the report line by line. If that run is wrong, no connector would have fixed it; if it is right, the connector becomes a convenience to price, not a leap of faith.

Anthropic lists a build-connector skill that can use Zapier where there is no direct connector. That is a route to investigate, not a guarantee that every software product can be linked safely or without a partner bill. Zapier's own plan, task allowance and connected-app permissions should be checked for the exact route. A sensible first plugin test avoids building a new connector at all: pick a skill that can use an account you already operate and can audit. If a custom link is essential, put its setup and ongoing task consumption in a separate budget line.

Keep the first run in approval mode

The marketplace says every workflow starts in approval mode. The Academy says nothing sends, posts or pays until the user says so. This is the right default for invoices, outreach, supplier messages and any money movement. It does not mean the staged result is necessarily correct. A human should compare at least one line with its source record, scan for items that do not belong and check the exact outward action. The reviewer should have authority for that action in the ordinary business process, not merely access to the Claude interface. The marketplace page also says that once you trust a workflow you can let it run on its own schedule, and the build-agent skill turns a weekly manual job into one you can run by name or on a schedule. Treat scheduling as a later decision, made per workflow after a run of clean reviews, rather than a setting to switch on during the trial.

Use the invoice-chase skill as an illustrative test. A four-person architecture practice has six invoices showing overdue in its export. One was paid yesterday but the accounting feed has not reconciled. Claude might rank all six and draft six reminders. The reviewer checks invoice identifiers, outstanding balances and payment dates against the accounting system before approving any email. The paid invoice is held back. The cost of the workflow includes this review time; the approval gate prevents a wrong reminder only if someone uses it.

A copyable instruction for the first run is: “Prepare an overdue-invoice list from the connected accounting account. For each item show invoice number, amount, due date and last recorded payment. Draft reminders only. Do not send, alter the ledger or propose a payment action. Mark any record with conflicting dates for manual review.” An illustrative output might list invoice 1042 at $1,800 with a due date 18 days ago, then mark it “check payment feed” because a note says a transfer arrived. The correction is to reconcile the payment, not to improve the wording of the reminder.

Do a second test with an outward action that has an obvious falsification check. A recruitment agency can ask for a call list from its CRM. If the list includes a candidate who opted out of contact, the team has a data or permission problem, even if Claude's summary is fluent. Review the contact record and the selection rule before sending anything. Do not treat “approval mode” as a replacement for the CRM's consent and suppression process. The owner should document what the reviewer must inspect on every run, then decide whether the volume justifies further automation.

Install only what the first task needs

Follow the Academy's route in Claude: the menu labelled Customize, then Plugins, Discover, Small Business and Add. Connect one source account for the first chosen skill. Run the skill by typing its slash command or describing the job in plain words; Anthropic says Claude can choose the relevant skill. If the wrong skill is selected, invoke the command explicitly. Start with a read-and-draft job whose answer you can verify from an existing report. Keep the first sample small and reversible.

For an illustrative architecture practice, the first job could be a Monday brief from a calendar and a project list, without payment or email actions. The owner compares two appointments and three deadlines with the originals. If a deadline is missing, inspect whether the account used by Claude could see that project. If the source was accessible but Claude missed it, narrow the request and run it again. This distinction matters: buying a higher plan will not fix a missing connector permission or an inconsistent project status.

Anthropic's Academy also says the plugin can be tailored to the business. After the first correct run, tell Claude what the practice does, what fields matter and how it should format a brief, then inspect the proposed plugin change before saving it. Keep a copy of the previous instructions and one representative output. A change that improves a weekly report can accidentally alter a related workflow; repeat the basic source check after customising. The shared Claude Projects tutorial is relevant if the team also needs common reference material, but the plugin should still have a named owner.

Where the apparent bargain becomes expensive

There are four budget leaks to watch. First, an annual seat price is paid up front, so the cash commitment is larger than the monthly equivalent implies. Second, a connected service may need a higher plan for a specific action. Third, an automation bridge may incur its own recurring usage. Fourth, a human must review staged actions and correct bad source records. None is a reason to avoid the plugin; each is a reason to put the real workflow on one page before purchase.

Review time is the leak most budgets leave blank, so put a number on it. Suppose the recruitment agency's owner runs invoice-chase every Monday on about twelve overdue invoices and spends two minutes checking each draft against the ledger: that is 24 minutes a week, or 1.6 hours across a four-week month. At an illustrative internal cost of $40 an hour, the check is worth about $64 a month, more than three times the $20 Pro seat. That is not a reason to skip the check; it is the honest cost line. If the drafts improve and the owner moves to spot-checking four of the twelve, the same sum falls to about $21, and that fall is the evidence the workflow is maturing.

Here is a filled-in illustrative pilot budget for a marketing agency. Claude Pro: $20 for one monthly seat. Existing CRM: already budgeted, no incremental charge assumed until verified. Design account: existing plan, with its ability to create the required assets still to be checked. Automation bridge: none in the first month. Review: one owner records actual minutes spent checking four draft outputs. The only confirmed new subscription is the $20 Claude seat. This is more honest than calling the whole workflow “$20 per month”, because the other cost cells are unresolved rather than zero.

Another mistake is treating a plugin catalogue as a set of fully working automations on day one. The marketplace describes many workflows, but each relies on usable source data and permissions. A cash-flow snapshot with unrecorded bills will not become reliable through a premium seat. A lead-finder workflow with an empty CRM will not know which accounts were already approached. Before upgrading, fix the missing records and decide who owns them. Use the workflow technology choice tutorial if you need to separate data repair from tool selection.

Set a stopping condition for the pilot as well as a success condition. For example, the marketing agency could decide that four consecutive staged campaign drafts must cite the correct approved brief, preserve the offer dates and require no more than routine wording edits. If two drafts invent a discount or draw on an outdated brief, pause the skill, inspect which source it read and correct the input before another run. Count the cost of these checks in staff time. A subscription can be cheap while a review-heavy workflow remains a poor buy. Conversely, one dependable weekly summary may justify a modest plan even if the agency never uses the other catalogue skills.

Check the first four weeks with evidence: which skill was run, which accounts it touched, whether the result matched the source, what had to be corrected, whether any approval was given and how much time the reviewer spent. Do not score a staged action as a completed business result. A drafted email is not an email sent; a proposed payment run is not a reconciled ledger. If one workflow repeatedly produces correct drafts and the reviewer remains comfortable, expand one connector or user at a time. If it fails, pause the workflow before adding more seats.

Decide when Pro is enough and when Team earns its cost

Pro is the straightforward entry for one owner testing the Small Business plugin. Team is justified when named colleagues need their own seats and the organisation needs shared administration and billing. Max and Ultra buy individual usage capacity; they are not mandatory plugin tiers. Enterprise is a separate security and purchasing decision with seat and usage terms that should be quoted for the organisation. For a solo owner, begin with the least costly eligible plan that supports the task and the required data handling; do not pay a Team minimum for a capability the Academy says Pro already includes.

At the end of the trial, answer five questions. Did the skill access only the intended accounts? Were source figures correct? Were proposed outward actions held for review? Did the owner spend less effort on a repeatable task after corrections? Did the live bill match the seat and partner-service budget? If any answer is unknown, the next step is another controlled run or a vendor-plan check, not a broader purchase. That makes the plugin's price a decision about a functioning process, rather than a purchase based on the number of skills in its catalogue.

Further reads

• Is Claude Team Worth It for a Small Business? — Decide whether shared administration warrants Team.
• How Much Does an AI Agent Cost a Small Business? — Budget for a broader automation project.
• How to Prepare Your Small Business for AI Agents — Set ownership and approval rules first.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Claude Academy, How to install and use the Claude for Small Business plugin; Claude marketplace, Small Business plugin page; Claude pricing page; Claude support, What is the Team plan? and Purchase and manage seats on Team plans (all checked 5 November 2026).