When should you use it?
- An agent needs data from a system that is not on its list of supported connectors
- Your ERP, maintenance or shop-floor software is old, on premises or has no public API
- Someone asked to connect the whole ERP and you need a scoped first step instead
Not for: Systems that already have a working connector in your account; Hosting or running infrastructure; the skill writes plans, specs and code, it does not deploy them; Sharing passwords or API keys in a chat; it refuses to take credentials that way
What do you need, and what do you get?
You need
- The vendor, product, version and hosting of the system, confirmed by its administrator
- One specific data need, such as closed work orders with actual cost, and how fresh it must be
- Access to the vendor's documentation and any registry or web search tools in your session
You get
- A short recommendation with the chosen access route for each system and what it depends on
- An honest connection status and three first questions, each assigned to the person who can answer
- A comparison of manual, export and API options with provisional effort
- A connection plan file recording every route checked, what was ruled out and a proposed field mapping
How does it work, step by step?
- Install the pluginAdd Claude for Manufacturing from the Claude plugin directory and finish its plant setup.
- Name the system preciselyGive the vendor, product, version and whether it is cloud or on premises. Ask the system's administrator if you are unsure.
- Scope one flowDescribe a single data need and the skill it unblocks, instead of asking to connect the whole system.
- Read the plan before buildingCheck which access routes it actually searched, the questions it assigns to your IT staff and the conditions still open.
- Hand credentials over safelyHave IT create a dedicated, read-only integration user and supply secrets through the environment, never in the chat.
The prompt
Our adapted version, written to work in Claude, Claude Cowork and Claude Code. Values in lime are placeholders.
Fill in the blanks (optional; the copied prompt uses your values)
We need {data_need} from {system_name} {version}, hosted {hosting}, refreshed {freshness}. Before proposing anything to build, check in order whether a vendor MCP server, a community server, a documented API, a read-only database view, a Zapier route or a scheduled export can deliver it. Tell me which checks you actually ran. Recommend the best workable option, list what still has to be confirmed and by whom, compare it with manual and export alternatives, and explain how it could break. Do not ask me for passwords or keys in this chat.
The original skill is published by anthropics on GitHub (anthropics/claude-plugins-official). Open it there to install or import it; the prompt above is our own adaptation.
What can go wrong?
- Skipping the search for an existing server. Vendors sometimes publish one quietly, and enabling it beats weeks of custom code.
- Taking a plan as proof that it works. Until your own records come back through the route, treat readiness and field mappings as unverified.
- Letting someone's personal login become the integration account. Use a dedicated user with the narrowest scope.
- Anthropic labels the plugin a reference implementation it does not actively maintain, so review its output with your IT lead.
Our take
Most small and mid-sized companies run at least one system that no AI connector supports, and the usual answer is either a big integration project or giving up. This skill's best move is refusing to start building until it has checked whether something already exists, in a fixed order that ends at a humble scheduled export. It also forces two habits that save real money: name the exact product and version, and scope a single data flow instead of the whole system. We like that it is blunt about honesty, too. It reports which checks actually ran, marks unchecked options as unchecked, labels effort as provisional and will not claim anything works until real records come back. Credentials stay out of the conversation, which should be the default everywhere. The trade-off is that its examples and target schema assume a factory, so other businesses will need to translate. In our dry run for a fictional distributor's on-premises ERP, it found no published server, settled on a read-only database view for closed orders and assigned three open questions to the people who can answer them.
Where the time goes back: Integration projects stall on vague goals and on building something that already existed. Checking ready-made options first and scoping one data flow turns a weeks-long discussion into a short plan your IT staff can act on.
FAQ
Is it only for factories?
Will it build and host a connector for me?
Who made it and what license does it use?
Source and credit
This playbook is based on work by anthropics. We summarize it and link to it and write our own adapted prompt.
- Original
- custom-connector skill (anthropics/claude-plugins-official, Claude for Manufacturing)
- Creator
- anthropics
- Published on
- GitHub (anthropics/claude-plugins-official)
- Original date
- Nov 12, 2026
- License
- Apache-2.0
- Originally for
- Claude
- AI dry run
- Nov 16, 2026: passed. Dry-run tested by our AI test agent on sample input. Ladder followed in order, honest about checks, credentials refused, plan before build.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Nov 16, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
The original is licensed under Apache-2.0 by anthropics. Modified: our adapted prompt and summary are our own wording; the original skill {"was unchanged at the source when we snapshotted it (live source later removed)" if src_removed(s) else "is unchanged at the source link"}.
Are you the creator? Claim or correct this page.