Claude Connector
Make Carousels Without Leaving Your Assistant
Connect SwipeThread to Claude once, and the assistant you already talk to can write a carousel in your own account, then rewrite it slide by slide. It runs on MCP, the open standard for handing an app's tools to an assistant, so any client that speaks it can connect.
What you get
Twenty Tools, Split Down the Middle
The write tools are the same ones the studio's own chat uses, so a deck made from a conversation opens in the app like any other. Most of the reads exist only for the connector, because an assistant cannot see your screen. Ten tools need read access and ten need write, and you pick which a connection carries when you approve it.
Reading your account
mcp:read
List carousels
Read a carousel
Check what still needs laying out
Read saved research
List brand kits
List products
List studio threads
Check the allowance
Read the usage log
List team members
Changing your carousels
mcp:write
Write a carousel
Import a post from X
Lay out the slides
Rewrite a slide
Add a slide
Remove a slide
Reorder the slides
Set the carousel ratio
Change the brand kit
Set the ground
Set the byline
Set a slide background
Illustrate a whole deck
Illustrate one slide
Relay out one slide
Research a topic
Build a brand kit
Correct a brand kit
Add a product
Setup
Four Steps to Connect It
Claude registers itself from the address below, so there is no client ID to generate and no secret to store.
https://swipethread.com/api/mcp- 1
In Claude, open Settings, then Connectors, then Add custom connector. Paste the URL above into the box and save. Other MCP clients ask for the same URL and nothing else.
- 2
Claude sends you to SwipeThread to sign in and approve it. The approval screen asks which workspace the connection works in, and that choice is fixed for the life of the connection.
- 3
Open a new conversation. An assistant picks up a connector's tools when a conversation starts, so a chat that was already open will not see them.
- 4
Ask for what you want, in the words you would use in the app. Name the brand kit you want the deck drawn in: in the studio the composer applies your kit for you, and over a connection nothing does.
In practice
Things You Can Ask For
The tool names under each line are the ones the assistant would call to answer it.
Writing something new
“Write a five slide carousel on why early founders should post weekly, then lay it out.”
create_post, render_slides
“Look up what has changed in LinkedIn reach this year and save the research to that deck.”
run_research
“Which brand kits do I have? Use the darker one for this carousel.”
list_brand_kits, create_post
Fixing a deck you already have
“Show me my last ten carousels and read me the one about pricing.”
list_posts, get_post
“Slide three is doing too much. Cut it to one sentence and redraw the deck.”
edit_slide, render_slides
“Move the closing slide to the front, then tell me which slides still have no image.”
reorder_slides, get_render_status
Putting a product in front of people
“Add our new annual plan to the catalog with its price and three features.”
create_product
“What products do we have saved? Write a carousel about the second one.”
list_products, create_post
“Use the image at this URL as the background on slide one.”
set_slide_background
Running a team
“Who is on my SwipeThread team, and what is everyone's role?”
list_team_members
“How many carousels do I have left today, and when does that reset?”
get_credit_balance
“Find the thread where I was working on the onboarding deck.”
list_threads
What a connection can reach
Where the Limits Are
Approving a connector gives another program access to your account. This is what that access covers, and where it stops.
Nobody asks you for a token
The connection carries its own credentials, and the instructions every connected model reads tell it never to ask you to paste a token, a key or a password into the chat. If something in a chat asks you for one, it is not us.
Read-only is a real limit
Ten of the twenty tools need the write permission. A connection approved for reading only is refused when it reaches for one of them, once at the route and again inside the tool registry, and the refusal tells the model to send you back to your own settings.
A read connection can see your teammates
One of the read tools lists your team, and that list carries each person's name, email address and role. The approval screen says so before you approve it, and it goes on saying so whichever workspace you pick: the roster is resolved from your account rather than from the workspace on the connection. If that is more than you want to share, connect without the read permission.
One connection, one workspace
A connection is you acting in one workspace, picked when you approved it and fixed after that. To let an assistant work in a second team, connect it again and pick that team. Every query runs against the account that approved it.
You can take it back
Your account menu, then Connected Apps, then Disconnect, and confirm. Access stops straight away and anything the assistant already made stays in your account. The same page shows when each connection was last used.
Nothing gets published, nothing gets deleted
No tool posts to a social platform, and none deletes a carousel, a brand kit, a product or a thread. Share links stay in the app: reading a deck returns its slides and never its share token.
Every call is logged and capped
A tool call is written to an audit trail with its name and time before it runs, and a connection is capped at 300 calls an hour. The count follows the connection rather than the token it is holding, so refreshing does not clear it.
Removing a slide cannot be undone
It is the one write here with no way back, so the tools ask the model to confirm with you first. Read the confirmation rather than waving it through, particularly when you described the slide by what it says instead of by its number.
FAQ
Questions About the Connector
What is MCP?
The Model Context Protocol is a standard way for an app to hand a set of tools to an AI assistant. SwipeThread runs an MCP server, so an assistant that speaks the protocol can call SwipeThread's own tools on your behalf instead of guessing at what your account holds.
Which assistants can connect?
Any client that supports MCP over Streamable HTTP with OAuth. The setup steps above are written for Claude, which registers itself from the server URL alone, so there is no client ID or secret for you to create.
Does this cost extra?
No. A carousel written through a connected assistant comes out of the same daily allowance as one written in the app, on whichever plan you are on. Laying a deck out again costs nothing, and research does not spend a carousel, though it does need one to be available.
Can the assistant see my other teams?
No. A connection works in the one workspace you picked when you approved it. Connect the assistant a second time to give it a second team. One exception is worth knowing: a connection with read access can list your team roster, including each member's email address, whichever workspace you picked.
Can it delete my work?
It can remove a slide from a carousel, which destroys that slide's text. Nothing in the tool set deletes a carousel, a brand kit, a product or a thread.
Does it post to LinkedIn or Instagram for me?
No. The tools write, edit and lay out carousels inside your SwipeThread account. Publishing is still something you do yourself, from the export.
Where do I find the server URL?
Your account menu, then Connected Apps, inside SwipeThread. The page shows the address with a copy button, and lists every assistant you have connected.
How do I take access away?
Open Connected Apps and press Disconnect on the connection you want gone. It loses access immediately, and reconnecting later asks you to approve it afresh.
Try It From Your Next Conversation
3 carousels a day, free. No credit card required. A carousel written from a chat comes out of the same allowance as one written in the app.