Setup guide

Connect your AI assistant to Syndic8

One address, your existing Syndic8 login, about five minutes. This page walks through the setup for each assistant, how to check it worked, what to do if it didn't, and what to ask first.

Connect your assistant

The short version

1
Add Syndic8 as a connector in your assistant. Paste one address, https://mcp.syndic8.io/mcp, into its connectors or integrations screen. Nothing to download and no key to store.
2
Sign in as yourself when the browser window opens, with the Syndic8 login you already have. The assistant gets your organizations and your permissions, nothing more.
3
Ask something. “What would Wayfair reject right now?” is a good first question. The assistant discovers the tools on its own.
connector settings · what you'll paste in
Name:     Syndic8
Server:   https://mcp.syndic8.io/mcp
Sign-in:  your Syndic8 login, in the browser window that opens

That's it. Your assistant lists the available
tools automatically once it connects.
Works with any assistant that supports remote MCP servers. One address for every organization; access follows your login.
Where you're working

Pick your assistant

The server address is the same everywhere: https://mcp.syndic8.io/mcp. Only the menu you paste it into changes. Claude and ChatGPT are self-serve; Copilot and Gemini go through your workspace admin.

Claude Cowork and the Claude desktop app

One click, from the Claude directory
  1. Open claude.ai/directory/syndic8 and choose Connect.
  2. Sign in with your Syndic8 login in the window that opens and approve the connection.
  3. In a new chat, enable the Syndic8 connector and ask your first question.

Prefer to add it by hand, or on a Team or Enterprise plan where the directory is restricted? Settings → Connectors → Add custom connector, then paste https://mcp.syndic8.io/mcp. Same sign-in, same result.

Claude Code

Self-serve
  1. In a terminal, run claude mcp add --transport http syndic8 https://mcp.syndic8.io/mcp.
  2. Start Claude Code and run /mcp to complete the sign-in.
  3. Ask it to list your Syndic8 organizations to confirm.

Same permissions, same tools; the only difference is that it lives in your terminal.

ChatGPT

Business, Enterprise and Edu plans
  1. Ask your workspace admin to enable connectors and Developer Mode if they aren't already on.
  2. Go to Settings → Apps & Connectors, turn on Developer Mode under Advanced settings, then choose Create.
  3. Name it Syndic8, paste https://mcp.syndic8.io/mcp, choose OAuth, and complete the Syndic8 sign-in when prompted.
  4. In the message composer, select the Syndic8 app before you ask; ChatGPT only calls a connector you've added to the conversation.

ChatGPT Pro can connect in read-only mode. Free and Plus plans can't add custom connectors yet.

Microsoft 365 Copilot

Admin setup
  1. A Microsoft 365 Global or AI administrator creates the connector in the admin center under Copilot → Connectors → Create a new connector → Connect to MCP server.
  2. They paste https://mcp.syndic8.io/mcp, choose OAuth 2.0, and register the Syndic8 sign-in in the Teams Developer Portal. Email support@syndic8.io and we'll send the OAuth details they need.
  3. They roll the connector out to the people who should have it. Each person signs in to Syndic8 with their own login the first time.
  4. Ask Copilot to use the Syndic8 connector in your question.

Microsoft's connector path currently exposes read tools. Bulk edits and other write tools are available through Claude, ChatGPT and Claude Code.

Google Gemini

Ultra plan or Gemini Enterprise
  1. In the Gemini app, open Settings → Connected apps → Other → Add a custom app.
  2. Paste https://mcp.syndic8.io/mcp. Sign in with your Syndic8 login when the browser opens.
  3. Enable the Syndic8 app and ask Gemini to use it.

Custom apps in the consumer Gemini app need Google AI Ultra. Gemini Enterprise customers add custom MCP servers through their admin console instead.

Cursor and other developer tools

Self-serve
  1. Add a remote MCP server with the address https://mcp.syndic8.io/mcp and transport HTTP.
  2. Complete the Syndic8 sign-in in the browser when the tool opens it.
  3. Anything that can add a remote MCP server works the same way; the steps only differ in where the setting lives.

No client ID, secret or API key to paste anywhere. If a tool asks for one, it's looking at the wrong field.

Confirm it worked

Three quick checks

1

Ask for your organizations

"List the Syndic8 organizations I can reach." A list back means the connection and the sign-in both worked. If you belong to several organizations, say which one you mean in each question.

2

Ask what you're allowed to do

"What am I permitted to do in [organization]?" The answer is your own permission set from Syndic8, so it also tells you which parts of this page apply to you.

3

Start read-only

A product count or a field profile is a safe first request. Save the first write for once you've seen the assistant return data you recognize.

If it doesn't connect

Four things it usually is

The assistant reports no Syndic8 tools

The connection wasn't completed, or the sign-in has expired. Remove the connector, add it again, and finish the browser sign-in.

The tools are there but every call is refused

You're signed in, but your account lacks those capabilities in that organization. Ask the assistant to list your permissions, then take the gap to your Syndic8 administrator. Nothing on the assistant side changes this.

An organization you expect is missing

Your account isn't attached to it in Syndic8. The server can only show what your login can reach.

Your assistant won't let you add a custom connector

Some workspaces restrict this centrally. Your IT or workspace administrator has to allow it first; the plan notes above say who that is for each assistant.

Still stuck? Email support@syndic8.io with the assistant you're using and what it said.

Before your IT team asks

What the assistant can see and change

Your permissions, nothing more

The assistant signs in as you and inherits your roles, organization membership and per-capability permissions from Syndic8. A refusal is final, not a prompt to try harder. Two colleagues connecting to the same address get two different sets of capabilities if their access differs.

Mostly read-only

41 of the 53 tools change nothing: queries, profiles, readiness checks and diagnostics. The 12 that write are gated by the same permissions as the app, and the assistant is told to say exactly what it intends to do before doing it.

Preview before write

Bulk updates have a read-only preview that reports the exact match count and current values. Deactivation needs an identity filter and an expected count, and aborts on a mismatch. Every bulk change returns an audit id.

No copy of your data

The assistant queries Syndic8 live on each request. Nothing is exported or cached on our side, and what it tells you is the current state of your catalog, not a snapshot.

Environments stay separate

A sandbox connection and a production connection are different connections. Demo and training work can't leak into live data by accident.

Disconnect any time

Remove Syndic8 from your assistant's connector settings and access ends. Your Syndic8 login and permissions are untouched.

Want the long version? The Syndic8 MCP Business Overview covers all 53 tools, the safety model and what MCP deliberately does not do. Ask us for a copy.

Prompt library

Things to ask once you're connected

Questions get further than commands. Scope to a collection when you can, name the channel you mean, and ask for the preview before any bulk change.

Readiness and rejections
“We go live on Wayfair in ten days. Run a readiness check on the launch collection and tell me what will fail, and why.”
“Item Weight is failing for 1,412 products on the Home Depot template. Diagnose the field and tell me how to fix it.”
“Run the image readiness check for our Amazon channel. Which products would be rejected on imagery alone?”
“Re-run the Target preflight on the Spring Outdoor collection and tell me if the failure count went down.”
Knowing your catalog
“What fields do we have for faucets, and what are the internal names?”
“Which attributes are least populated across the whole catalog?”
“What do we actually know about our outdoor lighting range? Give me fill rates by field.”
“Which product types are still active? Can I create products against this one?”
Templates and valid values
“We list on Amazon and Walmart. Where do the two templates disagree, and is that deliberate?”
“Compare our Wayfair template against the marketplace baseline. What no longer matches?”
“Which of our color values have no match in Lowe's controlled list?”
“Show me the valid values Home Depot accepts for Fixture Color Family.”
Fixing content
“Our brand name is spelled three ways across 3,000 SKUs. Show me the variants and preview a fix at organization scope.”
“Target wants different titles but nothing else should change. Preview a title update scoped to the Target destination only.”
“Rewrite the Amazon titles over 200 characters, keeping the brand first. Preview it before you write anything.”
“Which products in the Fall collection are missing a marketing description? Preview filling it from the short description.”
Images
“Home Depot says a third of our items have no imagery. Which uploads matched nothing?”
“Which products in the Fall assortment have fewer than three linked images?”
“Attach IMG_4471.jpg to SKU 88213-BLK; automatch missed it.”
Inventory, pricing and orders
“Which SKUs are out of stock at Zappos but in stock everywhere else?”
“Show me products with a Walmart price but no MSRP.”
“How much open purchase-order value do we have with Nordstrom this month?”
“Which products changed price in the last seven days, and on which channels?”
Honest boundaries

What the assistant does not do

  • It does not run exports or send data to retailers. Readiness checks are dry runs; syndication runs stay in the platform, on their schedules and connections.
  • It does not create or configure trading-partner connections. It can tell you a channel isn't connected; setting one up is app work.
  • It does not build templates from scratch. It reads, compares and diagnoses templates in depth; authoring is done in Template Admin.
  • It does not import files. Batch ingest creates products from structured records; file-based import flows remain a platform function.
  • It does not reactivate what has been deactivated. Reversal of a soft delete requires an administrator.
  • It does not replace the application. Visual review, approvals and the screens built for a specific job are still where that work belongs, which is why results carry links back into the app.

Rather have us connect it with you?

Book 20 minutes and we'll set it up on your own catalog, run the first readiness check together, and leave you with a prompt list for your channels.

Book a setup callWhat else the AI can do