FABX for AI · MCP connector · live · free

Every FABX tool, inside your AI agent.

Tell Claude Code, Cowork, Cursor or ChatGPT "compress this photo to 20 KB" or "merge these PDFs and Bates-number them" — the agent hands FABX the file and gets a link back. One connector, 15 tools, no tab-switching.

How to add it →
https://mcp.fabxlabs.com/mcp · free · no API key · works with Claude, ChatGPT and any MCP client
HereIn your browser, fully offline. Files never leave your device. Free, always.
In AIAdd one connector to Claude or ChatGPT. Say what you need; the file comes back done.

Set up in one minute

Claude Code · Cowork · Cursor · ChatGPT
Best fit · agents with file access

Claude Code, Cowork, Cursor, scripts

Add a remote MCP server (streamable HTTP) at https://mcp.fabxlabs.com/mcp. The agent reads your file, sends it to FABX, and returns a download link — fully hands-off. In Claude Code: claude mcp add --transport http fabx https://mcp.fabxlabs.com/mcp

ChatGPT

Settings → Apps & Connectors

Turn on Developer mode, click Create, paste the same URL, authentication None. Files you attach in chat are passed through by ChatGPT Agent; in plain chat you drop them on a one-time FABX link.

Claude.ai chat

Settings → Connectors → Add custom connector

Name it FABX, paste the URL, Add (Pro/Max/Team plans). Claude can't forward chat attachments to any connector, so it gives you a one-time link to drop the file on — then finishes the job. Honest note: for simple PDF merges Claude can also do it natively; FABX shines on exact-KB photo and signature targets.

How a request goes

3 steps
1 · Ask

Say it in plain words

"Compress this photo to 20 KB for my SSC form", "merge these three PDFs and Bates-number them", "make a UPI QR for ₹500". Your agent picks the right FABX tool.

2 · Hand over

Agent sends the file — or you drop it

Agents with file access (Claude Code, Cowork, Cursor) send the bytes directly. In a plain chat, FABX gives a one-time link; drop the file, say "done".

3 · Get

Download link comes back

Processed on FABX servers, held at most 1 hour, then deleted. Your files are never kept, sold or used for training.

Tools the connector exposes

one connector · 15 tools

Why one connector, not many

You add it once. Your AI sees every tool by name and chooses the right one for the job — image, PDF or QR. No account, no key, no permanent storage, no training on your files.

1
URL to add
15
tools inside it
0
files kept after 1 hour
₹0
free

Questions

What is an MCP connector?

Model Context Protocol is the open standard AI assistants use to call outside tools. Claude, ChatGPT, Cursor and many agents support it. A connector is just a URL you add once; after that the assistant can use the tools it provides.

Is it free?

Yes, for personal use, with fair-use limits (files up to 20 MB, links valid 1 hour). Paid keys for teams and high volume will come later; the free tier stays.

Are my files uploaded?

Unlike the website tools, the connector processes files on FABX servers (Cloudflare) so your AI can use them. Uploads and results live in temporary storage for at most 1 hour and are then deleted automatically; your files are never kept, sold or used for training. For Aadhaar, PAN, bank statements or anything sensitive, use the browser tools — they never upload.

Why does chat sometimes give me an upload link?

Chat apps (claude.ai, ChatGPT) keep your attachments inside their own sandbox and never pass the bytes to a connector — any connector. So FABX gives the assistant a one-time link; you drop the file, say "done", and it carries on. Agents that hold the file themselves (Claude Code, Cowork, Cursor, scripts) skip this entirely and send the file inline.

Which tools are worth it inside an AI?

The exact-size ones: photo under 20 / 50 / 100 KB with a minimum, signature 140×60 at 10–20 KB, passport 3.5×4.5 cm, PDF under 100 KB for a portal, Bates numbering with prefix and padding. Generic assistants guess at these; FABX hits the number.

Can I use it in my own app or script?

Yes — it is a standard MCP server over streamable HTTP at https://mcp.fabxlabs.com/mcp; any MCP SDK can call it. A plain REST API with keys is on the roadmap.