FRIDAI MCP and the Optimizer

Download, install and sign in

A regular Claude chat can connect to FRIDAI — you don’t need Claude Code. Even the free plan can add 1 custom connector.

Codex lives in the ChatGPT desktop app. ChatGPT’s Free, Go, Plus, Pro, Business, Edu and Enterprise plans can all use it; how much you can use depends on the plan.

1. Download Claude from the official site

   On your computer, open claude.com/download and choose the Mac or Windows version.

   Open the Claude download page

2. Install and sign in

   Open the downloaded file to install it. Open Claude from the Applications folder on a Mac or the Start menu on Windows, and sign in to your Claude account.

3. Prefer not to install anything?

   Open claude.ai in your browser, sign in and use the web version.

1. Download ChatGPT from the official site

   On your computer, open chatgpt.com/download and choose the Mac or Windows version. The Windows version installs from the Microsoft Store.

   Open the ChatGPT download page

   If you installed the Codex desktop app before, a normal update turns it into the new ChatGPT desktop app, and your existing chats and projects are kept.

2. Install and sign in

   Once it is installed, open ChatGPT and sign in to your ChatGPT account.

3. Switch to “Codex”

   Click the menu in the top-left corner and choose “Codex”.

Your first session

1. Say what you need in plain words

   Type in the message box and send it. Be clear about the result you want, for example “Check the Notebook for (brand name).”

2. It asks before using FRIDAI

   The first time it needs a FRIDAI tool, Claude asks for permission: “Allow once” allows just this time, “Always allow” stops it from asking again, and “Deny” blocks it.

3. New task, new chat

   Click “New” in the top-left corner to start a new chat.

1. Pick a folder

   When you start working, pick a folder; Codex uses the files in it. If you don’t have a suitable one, create a new folder on your desktop.

2. Say what you need in plain words

   Type in the message box and be clear about the result you want. Add any files you want it to look at to the chat.

3. It asks before it acts

   You can choose permissions below the message box. The official advice is to start with “Ask for approval”: Codex works inside the current folder and stops to ask you before going beyond it.

   “Full access” lets it change any file on your computer without asking you. Don’t turn it on casually.

4. New task, new chat

   Click “New chat” to start a new chat.

Before you start, check three things

• Your account has at least one active paid subscription. Free, trial, demo and deactivated accounts don’t show MCP Settings; if you just paid, refresh the dashboard and it will appear.
• You are signed in to Claude, on the web or in the desktop app; even the free plan can add 1 custom connector. No account yet? See Download and install.The ChatGPT desktop app is installed and signed in on your computer. Not yet? See Download and install.
• Authorization only ever happens on FansNetwork.ai, the official domain of the FRIDAI dashboard and MCP. If you see any other address, don’t sign in.

Connect in four steps

1. Open Claude’s connector settingsOpen Codex’s MCP settings

   In Claude’s left sidebar, click “Customize” → “Connectors”, then click “Add” in the top-right corner → “Add custom connector”.

   In the ChatGPT desktop app, open “Settings” → “MCP servers” and click “Add server”.

2. Paste the FRIDAI MCP URL

   Enter FRIDAI as the name, paste the URL below, and click “Continue”. Leave the settings on the next page as they are and click “Add”.

   Enter FRIDAI as the name, choose “Streamable HTTP” as the type, and paste the URL below. Once it is saved, click “Restart”.

   FRIDAI MCP URLCopy URL

   https://www.fansnetwork.ai/api/mcp/ruleops

   This is only the connection address; it contains no key.

   

3. Sign in to FRIDAI in your browser and check the request

   Click “Connect” and your browser opens FansNetwork.ai.

   The list shows that FRIDAI needs you to sign in. Click “Authenticate” and your browser opens FansNetwork.ai.

   After you sign in to FRIDAI, first check that the account and the number of brands on the screen are yours.

   

4. Back in the app, confirm it is connected

   Go back to Claude, start a new chat and ask it to “List the FRIDAI brands I can manage.” The first time, it asks whether it may use FRIDAI’s tools; click “Allow once” or “Always allow”.

   Go back to Codex and ask it to “List the FRIDAI brands I can manage.”

   If you can see your brands, you are connected.

   At no point do you need to see, copy or store an API key — and never paste any key into a chat.

Understanding the status in MCP Settings

You will find the connection status under “Current status” in “MCP Settings” in the dashboard’s left menu. It reflects this account’s authorization record, not a live check of each computer; to see whether a particular computer works, ask Claude or Codex on that computer to list your brands.

Authorization renews automatically, and reconnecting is quick

Each authorization is valid for 90 days. When it expires, the app automatically gets a new one, with nothing for you to do. If the app asks you to sign in again, or the permissions are incomplete, just sign in once more as below. This doesn’t change your Notebook, and you don’t need to subscribe again.

In Claude, go to “Customize” → “Connectors”, click FRIDAI, click “Connect” to reconnect, then confirm the authorization in your browser.

In the ChatGPT desktop app, go to “Settings” → “MCP servers”, find FRIDAI, click “Authenticate” to sign in again, then confirm the authorization in your browser.

Don’t want the app to connect anymore? In “MCP Settings”, open “Advanced setup and key management” and click “Revoke connection”. Claude or Codex loses access immediately.

What it can do for you once connected

FRIDAI MCP covers eight kinds of work. The first five work as soon as you connect; AutoPost and scheduled posts need one extra permission, “AutoPost settings” — here is how.

AI Notebook: read, update, roll back

• Read the Notebook, business details and Reply Settings that are currently live.
• Update the Notebook text directly, up to 50,000 characters. Before each update, the old content is always saved as a version.
• Switch the AI on or off, and edit your fixed business details: store name, address, phone, email, website, opening hours, products, other notes, and whether combined quotes are allowed.
• Link this Notebook to another of your Notebooks, or remove the link.
• See the last 50 versions and restore any of them; before restoring, the current content is backed up too.

If you edit the Notebook in the dashboard after it has read it, it refuses to overwrite your changes. Ask it to read the latest content and try again.

Reply Settings: most of what the dashboard offers, just by asking

The names below match “Reply Settings” in the dashboard. Just say “Set Comment Reply Delay to 10 seconds.” It changes only what you mention, and saves the previous values first.

Message Center: read-only

• List conversations, filtered by platform, date or keyword.
• Open a conversation to see, for each customer message, whether the AI replied — and if not, which check stopped it.
• Download a CSV backup. The download link is valid for 5 minutes, and permissions are checked again when you download.

It can’t reply to customers for you, and it can’t edit or delete messages. A conversation record saying “the AI wrote a reply” doesn’t necessarily mean it reached the customer.

AutoPost: see your settings and results, and change them too

For every change, it shows you before and after first, and applies it only when you say yes. AI-written drafts and carousels use your AutoPost quota. Carousel templates with photos are still set up in the dashboard, and for AI-avatar short videos you can email contact@fansnetwork.ai about our production service.

Scheduled posts: plan a whole batch at onceNew

These are the same scheduled posts as in AutoPost’s “Content Calendar” in the dashboard: posts scheduled in Claude or Codex show up in the dashboard, and posts scheduled in the dashboard show up for it too.

1. Tell it what to schedule

   For example, “One post each on Instagram and Threads at 8 pm every day next Monday to Friday,” then give it the text and images. You can schedule 1 to 30 posts at a time.

2. It shows you the whole batch first

   Each post’s date, time, platforms, opening words and images are listed. Problems are flagged right away — for example, an Instagram post with no image, an X post over the character limit, or a platform that needs reconnecting — and in that case nothing is saved.

3. Say yes, and the whole batch is saved

   You confirm the batch once. Either every post is saved or none is — never just half.

4. They post at the scheduled time

   Just like posts scheduled in the dashboard, they go out on time. Until then, you can ask it to edit or cancel any of them.

How to give it images and videos

• Already in the dashboard’s media library: ask it to “list the images in my media library” and pick from them.
• Available at a public link: paste the link and it saves the file to your media library. Google Drive and Dropbox share links work too — set them to “Anyone with the link can view”. Images must be 15 MB or less; videos must be MP4 or MOV and 100 MB or less. This step doesn’t publish anything, so there is nothing to confirm.
• On your computer: apps that can run commands on your computer, such as Claude Code and Codex, can upload them for you. A regular Claude chat can’t, so upload them to the dashboard’s media library first.

TikTok requires you to confirm privacy, interaction and content disclosure settings yourself on the posting screen, so please schedule TikTok posts in the dashboard. It also never posts immediately; the earliest it can schedule is 2 minutes from now.

To use AutoPost and scheduling, add “AutoPost settings” first

This permission is available when your account has an AutoPost subscription. Until you add it, AutoPost and scheduled posts are view-only; the Notebook, Reply Settings and Message Center are not affected.

1. Remove FRIDAI

   Open “MCP Settings” in the dashboard and, under “Optional: add the "AutoPost settings" permission”, click “Open Claude connector settings” — or go straight to “Customize” → “Connectors” in Claude. Remove FRIDAI.

   

2. Add it again with the same URL

   Follow the four steps above and add FRIDAI again with the same URL.

3. Confirm “AutoPost settings” on the authorization page

   The list on the FRIDAI authorization page now has an extra line, “AutoPost settings”. Check it, then click “Allow connection”.

   

1. Copy the permission command from the dashboard

   Open “MCP Settings” in the dashboard, find “Optional: add the "AutoPost settings" permission”, and click “Copy Codex permission command”.

   

2. Paste it into Codex and send

   Codex reconnects and your browser opens the FRIDAI authorization page.

3. Confirm “AutoPost settings” on the authorization page

   Check that “AutoPost settings” is in the list, then click “Allow connection”.

   

What it won’t do

• Message your customers, or edit or delete any message
• Post immediately: a scheduled post only goes out after you confirm it, at its scheduled time
• Schedule TikTok posts (please do that in the dashboard)
• Get your social platforms’ sign-in tokens
• Pay, change or cancel a subscription — and it can’t see your billing history either

Start the Optimizer

1. Nothing to install

   Once FRIDAI MCP is connected, the Optimizer is already there, and it always uses FRIDAI’s latest version.

2. Copy the “Improve AI support” line and paste it into a chat

   Once you are connected, “MCP Settings” shows “Connected! Copy one line and paste it into your AI”. Click “Copy” under “Improve AI support”, paste it into a Claude or Codex chat and send it, and it gets started.

   

Start it and pick the right mode

In the chat, just say “Use the FRIDAI Optimizer to…” followed by what you want it to do.

If one message asks for both, it handles them separately and never writes FRIDAI’s product information into your Notebook.

What a real report looks like

Below is an excerpt from a full diagnostic report the Optimizer produced for one brand. This Notebook covered two lines of business at once — medical devices and sports nutrition supplements — the highest regulatory risk there is. Brand, personal and organization names have been masked.

Excerpt from section 2: every item says what is wrong, the new-hire view, and how to fix it

Excerpt from section 7: customer questions run against the improved version

15 of 18 passed; 3 are blocked on [待補] items — not wrong answers, but correctly refusing to make things up.

How a full optimization runs

Once MCP is connected, you only need to say which brand to optimize — there is no need to paste your Notebook in yourself.

1. Read the current state

   It checks permissions, selects the brand you named, and reads the latest Notebook and Reply Settings.

2. Check-up

   It runs the 100-point check and lists every item that didn’t pass, anything that may contradict itself, and any questions still unconfirmed.

3. Sort out the facts, one question at a time

   Prices, offers, dates, opening hours, addresses, contact details, links, products, policies and rules for handing over to a human are each marked “confirmed”, “contradictory” or “missing”. If something contradicts, it asks you, one question at a time; it never picks one on its own or fills gaps with common sense.

4. Write the complete new version, up to 100 points

   All confirmed content and conditions are kept. Anything about how it runs — timing, probabilities, delays, schedules, skip lists, quiet hours and feature switches — goes into Reply Settings, not the Notebook. After rewriting, it runs the check-up again until it scores 100.

5. Make sure nothing from the original is lost

   The new version is compared with the old one item by item: nothing that was confirmed may be missing, and nothing you never provided may be added.

6. Conversation tests

   It runs two rounds of multi-turn conversation tests with AI-generated anonymous customers, then compares before and after across 20 scenarios. It doesn’t read real customer conversations, send messages, create bookings or publish anything.

7. Tests pass, update directly

   If you said “optimize” or “update”, it updates the Notebook directly once the tests meet the bar — no need to go back to the dashboard to approve. If you only said “check” or “give me a report”, you just get the report and your Notebook stays the same.

Pass criteria: miss one and it isn’t done

• A check-up score of 100, with no unresolved issues.
• Two rounds of conversation tests with 20 scenarios each. Each round has at least one three-turn conversation to confirm it remembers what was said earlier, and one question where the customer starts from a wrong assumption and gets correctly put right.
• Zero made-up information, zero pretending to have done something, zero privacy leaks, zero leaked internal instructions, and zero breaches of the policies you wrote down.
• At least 3 points on every scoring dimension (out of 5), and a weighted average of at least 4.
• The improved version must not perform worse than the original.

If the score isn’t 100, it tells you what is still missing instead of saying it’s finished.

Updating and rolling back: you decide

Once you have talked it through in Claude or Codex, it updates things for you directly — no need to go back to the dashboard to approve. Whether it changes anything depends on what you ask:

If someone else edited the Notebook in the meantime, it refuses to overwrite. Ask it to read the latest content and try again.

Ready-to-copy prompts

Start with a full check, then make smaller adjustments to tone or Reply Settings. Replace the words in brackets with your own.

The Optimizer

Official FRIDAI MCP prompts

AutoPost and scheduling

FAQ