Summary
Self-hosting can solve the ownership problem for you, but it doesn’t do anything for the interfaces you still have to navigate. Your documents are in one dashboard, your smart home runs from another, and before you know it, you’ve got a dozen Docker containers running. But you don’t have to use these apps yourself to get the most out of them. Turns out, you can wire Claude to your self-hosted stack, and get a lot done without ever having to deal with IP addresses, port numbers, and confusing interfaces. All you need is Claude, and an MCP server for the app you need to work with. n8n Ask Claude to run the workflow you forgot you built n8n is the strongest example here because its built-in MCP server supports more than retrieving information. Connected clients can search workflows, inspect those enabled for MCP, trigger and test them, and even create or edit workflows within their granted permissions (editing requires n8n 2.13.0 or later). That turns Claude into a natural language interface for your automation system, not a chatbot that spits out workflow JSON. Imagine opening a workflow that gathers updates from several feeds and produces a daily digest. Instead of opening n8n and navigating to that workflow, you can ask Claude to find and run it for you. The workflow still defines the actual process, Claude just offers a conversational entry point. Setup starts in the n8n settings under the Instance-level MCP option. Enable access, connect Claude using the provided server URL, and authorize it, preferably through OAuth. Existing workflows need explicit MCP exposure before the assistant can access their full data, run them, or modify them, although search can return previews of other workflows your user account can view. There’s also an MCP Server Trigger node for exposing tools within an individual workflow. Use that if you want a narrower interface on purpose rather than instance-level workflow management. Paperless-ngx Find the invoice you can’t name, and keep delete on a short leash A document archive is useless unless you can retrieve what you need when you need it. Plug in barryw’s PaperlessMCP and Paperless-ngx becomes your own document search engine with document search, metadata retrieval, uploads, tagging, and other management operations through MCP. This isn’t a built-in Paperless-ngx feature though, it’s an additional service you bolt on to the program using your instance URL and API token. Once Claude has access to your Paperless-ngx instance, you can make natural language requests to find specific documents with ease. Requests like finding invoices with a specific tag or a document that you remember the contents of but not the name can be handled easily. You can also ask Claude to propose relevant tags or titles before approving changes. The appeal isn’t to replace Paperless-ngx’s perfectly good document interface. It’s combining retrieval and housekeeping in one task. That said, broader requests should be treated cautiously as this bridge also supports deletion and bulk updates. Its documented safeguards include explicit confirmation for destructive operations and dry-run defaults for bulk operations, but those aren’t the reason you should let Claude run free through your document archive. Create the API token with only the access you need, since anything in your archive is text Claude will read. Home Assistant Start read-only, unless you want Claude flipping every switch Home Assistant is one of the best pieces of self-hosted, open source software you can find, and it becomes even more powerful when hooked up to Claude. Home Assistant’s official MCP server lets an AI assistant inspect your home’s state and control devices. Requests to turn on lights or control fans become answerable using live information rather than a screenshot or memory of whatever you told Claude. Home Assistant separates its capabilities into tool sets. Device control accesses entities exposed to voice assistants, while the broader Home Assistant set can read all entities, logs, and history and perform actions. The main MCP endpoint also has a read-only option, a much more sensible starting point if you’re skeptical about handing your entire smart house to Claude. Don’t expect the default connection to be narrowly restricted though. The current integration exposes all tool sets by default. It’s important you review permissions and access before authorizing Claude. Also, the documented custom-connector setup connects from Claude’s cloud, including when configured through Claude Desktop, so your instance needs secure remote accessibility. Local workarounds exist, but if they involve handing Claude a long-lived access token, that gives it unrestricted access to just about everything in your Home Assistant setup. Claude is an interface, not your new sysadmin Retrieval first, permissions second, dashboards forever The useful shift here isn’t that every self-hosted app needs AI. It’s that an assistant can become another interface to software that already runs on your infrastructure. MCP makes those capabilities accessible through conversation, and with permissions that deserve attention. Start with simple retrieval, keep the original dashboards close, and expand access only where it actually solves a problem, not just because you can. The goal isn’t to hand everything to Claude, it’s to spend less time navigating tools and more time using them.