Claude for Manufacturing: 17 Claude skills for the factory floor

On November 12, 2026, Anthropic released Claude for Manufacturing: a plugin with 17 skills and four connectors that turns a plant’s data into quotes, quality reports and requests to suppliers. According to Anthropic’s announcement, a person reviews and approves every one of those documents before it goes to a customer or a supplier.

What can Claude for Manufacturing do?

The skills cover sales, quality, operations, supply chain, design and setup. Some do a single job, others chain several steps together and stop wherever a human decision is needed. The announcement describes six of these workflows (it does not list all 17 skills):

1. For a new job, Claude reads the customer’s request, lists what is missing from it, compares the costs of similar past jobs and drafts a quote for the estimator.
2. After a failed quality inspection, it documents the defect, checks whether the same problem shows up in older reports and ranks the likely causes.
3. When a machine breaks down, it reads sensor exports, repair notes, photos and the manual, and the technician gets a list of causes to check.
4. When a supplier has to be replaced, it prepares the request for quotation and compares the bids that come back.
5. In Autodesk Fusion, it generates variants of an existing model, scores them against the goals the engineer set and drafts a decision memo.
6. For the plant manager, it puts together a production report: output against plan, machine downtime and what that downtime cost.

The data comes from four connectors. From Oracle NetSuite, Claude pulls orders, inventory levels, production steps and job costs. In Infor it checks purchase orders and stock, in Microsoft 365 it searches procedures, supplier emails and exported reports, and in Autodesk Fusion it works on the models. Systems without a connector can be hooked up too, by a roundabout route: you save regular exports from them to Microsoft 365, and Claude works on the files.

What is new here from an AI point of view?

After Claude for Legal, this is the next industry to get a ready-made toolkit from Anthropic instead of an empty chat window. This time it is a plugin, meaning a bundle of skills and connectors (in practice, MCP servers) that a team can adapt to its own forms and procedures and extend with its own skills.

Interestingly, the best picture of how Claude is supposed to work comes from the sample 8D report (an eight-step corrective action report) shown on the product page. A bore in the part should measure 12.000–12.025 mm in diameter, and 4 of the 8 measured pieces came out above the upper limit, the largest at 12.034 mm.

Claude does not name a culprit in it. The causes section opens with the words “ROOT CAUSE NOT VALIDATED”, followed by three hypotheses with the evidence for and against each: a wrong offset after an insert change, a gauge that reads high, and gradual drift in the dimension. The second hypothesis comes with a note that a constant 0.009 mm over-read would bring all eight readings back within tolerance, so the measurements alone cannot rule out a gauge error.

Next comes a proposal for the cheapest test that settles the question: check the gauge against a certified reference and measure the same pieces again. The corrective actions are marked “PROPOSED, NOT CHOSEN”, with no owners and no deadlines.

Does Claude control the machines?

According to the product page, Claude only produces documents: it does not operate equipment, does not run production control and writes nothing to the systems it reads from.

Changes in a system only become possible with a separate plugin from Infor. In the example from the announcement, a steel delivery slips by two weeks, the buyer asks Claude which customer orders are affected and how much steel other plants have, and once the buyer approves, Claude places an order with another supplier and reschedules production in Infor (within the permissions the buyer has in Infor).

Operating hardware is the job of a different Anthropic project, MHS (Model Hardware Standard), which I wrote about when covering the AI for Science program. It is a specification meant to let AI agents operate physical equipment safely, for now in a limited research preview for labs and selected manufacturers.

What should you watch out for?

First, the plans. The announcement mentions the Team and Enterprise plans, while the FAQ on the product page says “paid plans”, including Enterprise bought through AWS Marketplace. An admin installs the plugin from the plugin directory, and after that it is enough to type “Set us up”: Claude asks which systems the team uses, tests the connectors and lists the skills that will not fully work yet.

Second, the type of production. According to the FAQ, the plugin fits best where every job has its own quote, drawing and routing, and quality problems are written up in NCRs (non-conformance reports) and 8D reports. Process manufacturers (food, chemicals, steel) will for now mostly benefit from the morning production report, breakdown diagnosis and the supplier skills.

Third, control over changes. On the Team and Enterprise plans, the owner can require approval before a connector changes anything, or block changes altogether. On Enterprise, an admin can also make the plugin available to a single group only.

Anthropic will show the plugin in action in a webinar on November 17 at 11 a.m. Eastern Time. The agenda also includes a first look at MHS.

FAQ

How much does Claude for Manufacturing cost?

Neither the announcement nor the product page gives a separate price for the plugin. They only name the plans it can be installed on.

Does Claude for Manufacturing replace an ERP or quoting software?

According to the FAQ on the product page, it does not. Claude drafts a quote from the cost history, routings and earlier jobs in the ERP, and if the plant has separate estimating software, it can work from that tool’s exports.