
Fluf
io.flufPublished by fluf.iov0.2.7Updated Oct 24, 2026
Manage a FLUF resale account from an AI agent: inventory, crosslisting and unified orders.
Overview
Lets an assistant manage a FLUF resale account: read inventory and orders, crosslist products to marketplaces, fix failed listings, and file support tickets.
- What you can do
- Exposes six tools for a FLUF Connect resale account: list_channels shows which marketplaces are connected, list_products shows products and where they are live, crosslist lists products on marketplaces, get_orders returns orders across all connected marketplaces, list_drafts and approve_draft handle listings a marketplace refused, ask_intesa queries FLUF's own assistant, and report_bug raises a support ticket. It also ships named recipes as MCP prompts, such as dead_stock, morning_sales, listing_failures, expansion and support, which chain the tools into workflows.
- When to use it
- Use it when you sell on multiple marketplaces through FLUF Connect and want an assistant to research what to crosslist, summarize sales, diagnose failed listings, or open support tickets without handling per-marketplace APIs or credentials.
- Requirements
- A FLUF Connect account on an active paid plan, plus a personal access token supplied as the FLUF_API_TOKEN environment variable. The recommended Claude Desktop bundle needs no Node install; the npx route requires Node 18 or later. FLUF_BASE_URL optionally overrides the API host, which defaults to the FLUF site.
Installation
In SourceWeft
- Open Fluf in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
fluf-mcp — FLUF Connect for AI agents
A Model Context Protocol server that lets an AI agent (Claude, Cursor, Windsurf, Cline, and anything else that speaks MCP) work a FLUF Connect account: read inventory, list items across marketplaces, read orders, ask FLUF's own assistant a question, and raise a support ticket.
One account, one token, every marketplace you've connected — the agent never has to learn a per-marketplace API, and never handles your marketplace credentials.
Six tools are the primitives. Five recipes are the outcomes — named workflows that chain the tools in the right order, surfaced in your client as slash commands. Most people want a recipe.
Recipes
Recipes ship as MCP prompts. In Claude Desktop and Claude Code they appear as slash commands; in Cursor and other clients, in the prompt picker.
Every argument is optional unless marked, and every recipe has a sensible
default — /dead_stock on its own works.
Two things the recipes are deliberately strict about, because getting them wrong is how an agent loses a seller's trust:
- Nothing gets listed without confirmation.
dead_stockandexpansionresearch and recommend; they callcrosslistonly after you say yes. - Queued is reported as queued. Extension-first channels finish in your own browser, so an item can be accepted but not yet live. The recipes say so rather than claiming success, and they don't re-push (a repeat can duplicate the listing).
Tools
Get a token
Create one at https://fluf.io/connect/developers — in FLUF Connect that's More → Developers in the left sidebar. It's shown once, so store it like a password. Revoke it on the same screen any time.
You need a FLUF Connect account on an active plan; the API is a paid feature.
Install
Two ways in. Pick the first unless you have a reason not to.
Claude Desktop — one click, no Node (recommended)
Download fluf-mcp-<version>.mcpb from the
latest release and open it.
Claude Desktop shows an install dialog, asks for your token in a form field, and
that's it — no terminal, no JSON, and no Node install: Claude Desktop ships
its own Node runtime and uses it for installed bundles.
Everything else — npx
For Cursor, Claude Code, Windsurf, Cline, or if you'd rather manage it yourself.
This route does need Node 18+ on your machine — the
runtime bundled with Claude Desktop is only used for installed .mcpb bundles,
not for claude_desktop_config.json entries, which spawn command from your
system PATH.
Configure
Only needed for the npx route — the .mcpb bundle configures itself.
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json:
Claude Code
Cursor
~/.cursor/mcp.json (or a per-project .cursor/mcp.json):
Environment variables
Using it
Reach for a recipe when you want an outcome — they encode the tool order and the caveats. Otherwise ask in plain language and the agent picks the tools:
"What have I got in stock that isn't on eBay yet? List the ten cheapest."
"Show me everything that sold last week and which channel it sold on."
"Why did my last five Vinted listings fail?"
Three things worth knowing:
- Always call
list_channelsfirst. The available marketplaces differ per account and change over time; don't hardcode a list. crosslistis not always instant. Some channels are handed to your own browser session to complete, so the response may say queued rather than listed. Read the per-channel status; don't assume success.ask_intesais the slow, clever one. It hands the question to Intesa, the assistant inside FLUF, which runs its own multi-step investigation before answering — so it can explain why something happened, not just report what is. Replies can take up to a minute. Use the direct tools for simple reads; reach for this when the question is diagnostic or open-ended.
Develop
Smoke-test without an MCP client:
Recipes live in src/prompts.ts — one entry in RECIPES per
recipe, no registration step. They're plain text aimed at the agent, so editing
one needs no client change.
Support
License
MIT — see LICENSE. © FLUF.io.
The server itself is open; the account it talks to is not. You'll still need a FLUF Connect account on an active plan for any of it to do anything.
Source: README.md at commit a121d67
Tools
0Version history
2- v0.2.7LatestSep 27, 2026
- v0.1.2Sep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts