Why I Got Tired of Copy-Pasting Files Into Claude
There's a specific kind of frustration I remember from early on: I had a folder of markdown notes I wanted Claude to help me reorganize. So I'd open a file, copy the contents, paste it into Claude, get a response, copy the response, open the file again, paste it back in, save it. Then repeat. For twelve files.
By file four I was ready to just do it manually. That whole workflow is backwards — I'm doing all the tedious work and Claude is just sitting there waiting to be fed content like a very smart bird.
What I actually wanted was for Claude to just... open the folder and handle it. And it turns out that's exactly what the MCP filesystem server lets you do. Once I got it set up, Claude could read files, write updated versions back to disk, and work through a whole directory while I went and made coffee. That's the version of AI assistance I signed up for.
In this article I'll walk you through setting up MCP filesystem access from scratch — what it is, how to configure it, and a few practical examples of what you can actually do with it once it's running.
Quick Recap: What MCP Actually Is
If you've read the MCP vs Plugins explainer on this site, you can skip ahead. But if not — MCP stands for Model Context Protocol. It's an open standard that lets AI models like Claude connect to external tools and data sources through a standardized interface.
Think of it like a USB standard. Before USB, every device had its own proprietary connector. MCP does for AI tools what USB did for peripherals: one protocol, many devices. The filesystem server is just one of those devices — it gives Claude a structured way to read directories, open files, and write content back to disk on your local machine.
Crucially, you control exactly which folders Claude can access. It's not just roaming around your whole hard drive — you define the allowed paths when you configure the server, which is the part that makes this feel safe to actually use.
What You'll Need Before Starting
This is genuinely an intermediate setup — not because it's complicated, but because it requires a few things already in place. Here's the checklist:
1. Claude Desktop app. MCP servers run as local processes that Claude Desktop connects to. The web version of Claude at claude.ai doesn't support MCP (yet). If you don't have Claude Desktop, grab it from anthropic.com/claude — it's free to download.
2. Node.js installed. The official filesystem MCP server is a Node.js package. Run this in your terminal to check if you have it:
node --version
→ v20.11.0 (anything v18+ is fine)If you get a "command not found" error, head to nodejs.org and install the LTS version. It takes about two minutes.
3. A folder you want Claude to work with. Pick something low-stakes for your first test — a folder of notes, a small project directory, something where it won't matter if things get weird while you're learning the ropes.
Setting Up the Filesystem MCP Server
Anthropic maintains an official MCP filesystem server as part of their model-context-protocol GitHub repo. You don't need to clone it or build it — you can run it directly via npx, which means Node will download and run it on demand.
The configuration all happens in a single JSON file that Claude Desktop reads on startup. Let's find it and edit it.
On macOS: The config file lives at ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: It's at %APPDATA%\Claude\claude_desktop_config.json
Open that file in any text editor (VS Code works great). If it's empty or doesn't exist yet, that's fine — you're creating or replacing the whole thing. Here's the config you want:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"/Users/yourname/Documents/my-notes"
]
}
}
}Replace /Users/yourname/Documents/my-notes with the actual path to the folder you want Claude to access. You can add multiple paths as additional arguments in that array — just add more strings to the list.
Start with one specific folder
Don't be tempted to just pass your entire home directory on the first run. Give Claude access to exactly the folder you need for your current task. You can always add more paths later — it's much easier to expand access than to wish you'd been more careful.
Save the config file, then fully quit and relaunch Claude Desktop. Not just close the window — actually quit the app so it restarts fresh and loads the new config.
Verifying It Actually Worked
When Claude Desktop loads with a filesystem server configured, you should see a small hammer icon (🔨) in the bottom left of the chat input area. That's the MCP tools indicator. Click it and you should see filesystem tools listed — things like read_file, write_file, list_directory, and a few others.
If the icon isn't there, the most common culprits are a JSON syntax error in the config file (missing comma, extra bracket) or a path that doesn't actually exist on your system. Double-check both.
To do a quick sanity test, just ask Claude: "Can you list the files in my allowed directory?" It should call the list_directory tool and show you what's in the folder. When I first got this working and saw it actually list my real files back at me, I had a little moment. It felt like a genuinely different kind of AI interaction.
Three Things You Can Actually Do With This
Okay, the setup is done. Let's talk about what this is actually useful for, because "Claude can read files" is only interesting if you know what to ask it to do.
1. Batch-edit a folder of markdown notes. This is the use case that got me hooked. I had a bunch of blog draft files with inconsistent formatting — some used # for titles, some used ##, some had front matter, some didn't. I asked Claude: "Read all the .md files in my drafts folder, standardize the heading structure so every file starts with a single # title, and write the updated versions back to the same files." It worked through them one at a time, showed me each change, and saved them all. Would have taken me an hour manually.
2. Generate a summary or index of a project directory. Useful when you've inherited someone else's code or notes and want to get oriented fast. Try: "List all the files in this directory, read each one, and write a new file called README.md that summarizes what's in each file in two sentences."
3. Review and annotate code files. Point Claude at a folder of Python or JavaScript files and ask it to read through each one and add comments explaining what each function does, then write the annotated versions back to disk. This pairs really well with the AI code review concepts from other articles on this site.
Always keep a backup before write operations
Before you ask Claude to write files back to disk, make sure you either have a git repo tracking changes or a manual copy of the folder somewhere. Claude is generally careful, but write operations are real and immediate — there's no undo button in the filesystem.
A Note on What Claude Will (and Won't) Do
One thing that surprised me: Claude will sometimes pause and ask for confirmation before doing a write operation, especially if it's about to modify a bunch of files. That's actually a good thing — it's the model being cautious rather than just blasting through your files. You can tell it to proceed, or you can review the proposed changes first.
Claude also won't access any paths outside the ones you've explicitly listed in the config. If you gave it access to /Users/yourname/notes, it genuinely cannot read /Users/yourname/Documents — the server enforces those boundaries. This is one of the things that makes MCP feel more trustworthy than just giving an AI a terminal and hoping for the best.
Where to Go From Here
Once you've got filesystem access working, the natural next step is looking at what other MCP servers exist. The same config file can hold multiple server entries — you could have filesystem access running alongside a database server or a GitHub integration. Check out the Useful MCP Servers article on this site for a good tour of what's available.
You can also start experimenting with writing your own MCP server if you want to connect Claude to something more custom — but that's a whole separate rabbit hole. The Building Your First MCP Server from Scratch article covers that if you're ready for it.
For now though, just getting Claude reading and writing your local files is already a genuine productivity unlock. The copy-paste marathon is officially over.
More tutorials in this category, or explore the full field guide.