
filester-mcp
io.github.coltonbearden/filester-mcp
Unofficial MCP server for the Filester (filester.me) API: uploads, files, folders, account.
Documentation
filester-mcp
An MCP server and a command-line tool for the whole Filester file-hosting API. Upload files (streamed, up to the API's 10 GB), organise them in folders, move and delete them, and read file details and account usage, from Claude, Cursor, VS Code or any other MCP client, or from a shell script.
Unofficial. This project is not made by, endorsed by or affiliated with Filester. It is for filester.me, not for other services that share the name (filester.net, filester.app and others). It talks to Filester's public REST API with your own key.
- Every endpoint is a tool: all 16 of the API's endpoints, with every parameter it documents.
- Responses come back as the API sends them, with nothing reshaped or dropped.
- Uploads stream from disk, so a multi-gigabyte file never sits in memory. Progress is reported to clients that ask for it.
- Rate limits are handled. A
429waits outRetry-Afterand tries again, and the rate-limit headers are passed back. - A key is optional. Without one you get anonymous guest uploads and the health check.
Get a key
Sign in at filester.me and copy the API key from your Account Settings page. Treat it like a password: generating a new key cancels the old one at once.
Install
You need Node.js 20 or newer. Each install line below runs the latest version from npm. Replace
your-key with your key, or leave the key out to use guest uploads only.
Claude Code
claude mcp add filester --scope user -e FILESTER_API_KEY=your-key -- npx -y filester-mcp
Claude Desktop
The one-click way: download filester-mcp-<version>.mcpb from the
latest release and open it. Claude Desktop shows an
install dialog and asks for your key.
Or add this to claude_desktop_config.json (Settings → Developer → Edit Config):
{
"mcpServers": {
"filester": {
"command": "npx",
"args": ["-y", "filester-mcp"],
"env": { "FILESTER_API_KEY": "your-key" }
}
}
}
Cursor
Add to ~/.cursor/mcp.json (every project) or .cursor/mcp.json (one project):
{
"mcpServers": {
"filester": {
"command": "npx",
"args": ["-y", "filester-mcp"],
"env": { "FILESTER_API_KEY": "your-key" }
}
}
}
VS Code
Add to .vscode/mcp.json. VS Code asks for the key once and stores it for you:
{
"inputs": [
{ "type": "promptString", "id": "filester-key", "description": "Filester API key", "password": true }
],
"servers": {
"filester": {
"type": "stdio",
"command": "npx",
"args": ["-y", "filester-mcp"],
"env": { "FILESTER_API_KEY": "${input:filester-key}" }
}
}
}
Any other MCP client
The command is npx -y filester-mcp, which serves MCP over stdio. The key goes in the environment as
FILESTER_API_KEY.
Streamable HTTP (local)
For a client that connects to a URL instead of starting a command:
npx -y filester-mcp --http # http://127.0.0.1:8723/mcp (--port to change it)
- Stateless: each request gets a fresh server.
- Keys per request: a request may bring its own key as
Authorization: Bearer <your-key>. Otherwise the server'sFILESTER_API_KEYis used, if it has one. - Loopback only: it listens on 127.0.0.1 and refuses requests whose
HostorOriginisn't local. That's deliberate. The upload tools read paths on the machine the server runs on, so a server other machines could reach would let them upload its files.
Configuration
| Variable | What it does |
|---|---|
FILESTER_API_KEY | Your API key. Without it, only guest uploads and health work, and every other tool says it needs a key. |
FILESTER_API_KEY_FILE | A file that holds the key, read when FILESTER_API_KEY isn't set. The flag --api-key-file <path> does the same. On Linux and macOS it warns if the file can be read by other users. |
FILESTER_BASE_URL | The API's address, if Filester ever moves it. Default https://u1.filester.me. |
FILESTER_MAX_RETRY_WAIT | How many seconds one call may spend waiting out 429s before it returns the 429 to you. Default 300. |
Tools
Each tool returns the API's JSON body exactly as sent, as text and as structured content. When the API sends
rate-limit headers (it does on uploads), they come back too, as a second block {"rate_limit": {...}}. An API error
comes back marked as an error, with its HTTP status and body.
| Tool | What it does | Example arguments |
|---|---|---|
upload_file | Upload a local file (streamed). Returns url, slug, file_id. Without a key, or with guest: true, it's an anonymous upload. | {"path": "/home/me/video.mp4", "folder": "a1b2c3d4e5f6g7h8"} |
get_upload_status | Processing status of an upload. Needs no key. | {"slug": "aBcDeFg"} |
list_files | Your files, a page at a time. Filter by folder (or "root") or by name. | {"per_page": 50, "search": "invoice"} |
get_file | One file's details, views and downloads, by slug, UUID or numeric id. | {"id": "aBcDeFg"} |
move_file | Move a file into a folder, or back to the top level. | {"file": "aBcDeFg", "folder": "root"} |
move_files | Move up to 200 files at once. Each result is reported separately. | {"files": ["aBcDeFg", "xYzAbCd"], "folder": "a1b2c3d4e5f6g7h8"} |
delete_files | Delete files for good. | {"identifiers": ["aBcDeFg"]} |
list_folders | Every folder, with parent, to rebuild the tree. | {} |
create_folder | A folder or subfolder, public (default) or private, optionally with a password. | {"name": "Episodes", "parent": "a1b2c3d4e5f6g7h8", "public": false} |
list_folder_files | The files in one folder. | {"identifier": "a1b2c3d4e5f6g7h8"} |
set_folder_thumbnail | Upload an image (JPG, PNG, GIF or WebP, up to 5 MB) as a folder's cover. | {"folder": "a1b2c3d4e5f6g7h8", "path": "./cover.jpg"} |
move_folder | Move a folder and everything in it under another parent. | {"folder": "9f8e7d6c5b4a3210", "parent": "root"} |
move_folders | Move up to 200 folders at once. | {"folders": ["9f8e7d6c5b4a3210"], "parent": "a1b2c3d4e5f6g7h8"} |
delete_folders | Delete folders and the files directly in them. Subfolders move to the top level. | {"identifiers": ["a1b2c3d4e5f6g7h8"]} |
get_account | Storage used and limit, file and folder counts, API requests today. | {} |
health | Whether the API is up. Needs no key. | {} |
What the API itself allows
These limits are Filester's own; this server adds none of its own.
- Size and storage: 10 GB per file, and 10 GB of storage per account.
- Rate limit: 1,000 requests an hour per key.
- Folders: names up to 100 characters, nested as deep as you like.
- Expiry: files with no views or downloads are deleted after a while. The API docs say 45 days; the FAQ says 35.
- No download endpoint: a file is shared as its page,
https://filester.me/d/<slug>(theurlfield). Files in a private folder aren't reachable there (403). - No rename, no settings changes: the API can't rename files or folders, or change a folder's public setting or password after it's created.
- Guest uploads are permanent: they can never be deleted or managed.
- Deletes can't be undone. Deleting a folder deletes the files directly inside it. Its subfolders aren't deleted; they move to the top level.
docs/api-notes.md has what this project found when it tested the live API, dated, including the cases the docs leave open.
Large uploads
Uploads stream from disk, so file size doesn't change the server's memory: in testing, a 1 GiB upload peaked at 131 MiB of resident memory, against 119 MiB for a 50 MiB one.
- Progress: an MCP client that asks for progress gets a notification about twice a second.
- Client timeouts: a long upload can outlast an MCP client's own per-call timeout. In Claude Code, raise
MCP_TOOL_TIMEOUT(milliseconds). - Very large files: these are often easier from the command line:
filester upload big.mkv.
Command line
The same package installs a filester command over the same client, for scripts, cron jobs and tools like n8n:
npx -p filester-mcp filester health
export FILESTER_API_KEY=your-key
filester upload ./report.pdf --folder a1b2c3d4e5f6g7h8
filester files --per-page 100 --search report --json | jq '.data[].url'
echo 'secret' | filester create-folder Private --private --password-stdin
filester delete-files aBcDeFg xYzAbCd
- Output: every command prints the API's response.
--jsonprints the body exactly as it came, for scripts. - Exit codes:
0done,1the API answered with an error,2usage,3needs a key,4couldn't reach the API or read a file. - Help:
filester --helplists every command.
Security
- Where the key goes: only to Filester's API, as a Bearer token. It's never logged, and the server writes nothing to disk.
- Upload paths:
upload_fileandset_folder_thumbnailread a path on the machine the server runs on. Anything that can call the tools can upload any file that user can read. Run the server only for clients you trust. - Reporting: see SECURITY.md to report a vulnerability.
Development
npm ci
npm run build # dist/
npm test # unit tests against a local fake of the API (no network)
npm run test:slow # an upload the fake reads for over 5 minutes
FILESTER_API_KEY=… npm run test:live # the live suite, in a folder of its own, deleted afterwards
npm run lint && npm run typecheck
- Releases are automatic: semantic-release reads the commit messages (Conventional Commits) on
main. See docs/RELEASING.md. - Decisions are in docs/DECISIONS.md.
Licence
MIT. "Filester" is the name of its owners' service, and this project uses it only to say what it works with.
filester-mcpnpm install filester-mcp