Filesystem MCP server in n8n file access for workflow agents.
n8n's MCP Client Tool node connects over SSE or streamable HTTP — and mcp-server-filesystem speaks stdio. Here's how to bridge the two and give your n8n AI Agent read, write, and search access to local files.
mcp-server-filesystem (from the ByteDance UI-TARS repo, github.com/bytedance/ui-tars-desktop, 36.3k stars) gives an agent filesystem access: reading, writing, and searching local files. In n8n, that means an AI Agent node that can pull a report off disk, write generated output to a folder, or search a directory tree as a workflow step.
There's no config file to edit in n8n. MCP servers are attached through the MCP Client Tool node, which you drop into a workflow and point at an SSE or streamable HTTP endpoint. The one wrinkle: this npm server is stdio-only, so it needs a small HTTP bridge in front of it.
The transport gap, and how to close it
n8n's native MCP Client Tool node expects a network endpoint. mcp-server-filesystem launches as a local subprocess and talks over stdin/stdout. The standard fix is a stdio-to-SSE gateway — supergateway is a common choice — that launches the server as a child process and exposes it on a local port.
Run the bridge on the same machine as your n8n instance. If n8n runs in Docker, the bridge (and the files you want to expose) must be reachable from inside the container — host.docker.internal instead of localhost on Mac and Windows.
Start the bridge, then configure the node
Start the gateway with the filesystem server's real package name and the directory you want to expose, then paste the resulting endpoint into the MCP Client Tool node. The trailing path scopes what the agent can reach — consult the README in the UI-TARS repo for additional flags.
# mcp-server-filesystem is stdio-only; expose it over SSE:
npx -y supergateway \
--stdio "npx -y mcp-server-filesystem /data/shared" \
--port 8000
# In n8n: add an "MCP Client Tool" node to your AI Agent
# SSE Endpoint: http://localhost:8000/sse
# (use http://host.docker.internal:8000/sse if n8n runs in Docker)Wire it into an AI Agent workflow
Attach the MCP Client Tool node to an AI Agent node the same way you'd attach any other tool. Once connected, the agent sees the filesystem operations in its tool list and decides when to call them based on the task in your prompt. You can restrict which of the server's tools are exposed in the node's settings — useful if you want read-only behavior in a workflow that shouldn't write anything.
Troubleshooting in n8n
No tools showing up in the agent? Execute the workflow once or click the node's connection test — n8n connects lazily, so a freshly added node may simply not have dialed out yet. If the connection fails outright, verify the bridge process is still running and that the endpoint URL includes the right path for your gateway.
The most common failure is network scoping: n8n Cloud cannot reach localhost on your laptop, full stop. A locally bridged stdio server only works with self-hosted n8n on the same machine or network. Second most common: the directory passed to the server doesn't exist inside the environment where the bridge runs, so the server exits immediately and the endpoint goes dead. Restarting n8n isn't usually required — fixing the bridge and re-executing the workflow is.
See it on the marketplace
The server's live listing at https://loomal.ai/marketplace/mcp-server-filesystem shows its probed tool list, so you can check what your n8n agent will see before you build the workflow. Maintainers can claim listings like this one on Loomal to publish tool lists and set per-call pricing for paying agents.
FAQ
How do I install the Filesystem server in n8n?
Run the npm package behind a stdio-to-SSE gateway (for example supergateway), then add an MCP Client Tool node to your workflow and point its endpoint at the gateway's URL. The node attaches to an AI Agent node like any other tool.
Where is n8n's MCP config file?
n8n doesn't use one. MCP servers are configured per-workflow through the MCP Client Tool node's UI — endpoint URL and optional credentials — rather than a JSON file on disk.
Why can't n8n Cloud see my filesystem server?
Because the server and its bridge run on your machine, and n8n Cloud's servers can't reach your localhost. Use self-hosted n8n on the same machine or network, or expose the bridge through a tunnel you control — bearing in mind you'd be exposing file access to whatever can reach that URL.
What if the tools disappear after working once?
The bridge or the underlying server process died. Check the terminal running supergateway for a crash — usually a bad directory argument or an npx cache issue — restart it, and re-execute the workflow. n8n itself rarely needs a restart.
Browse more MCP servers for n8n.
Find endpoint-ready servers with live-probed tool lists.