Your assistant can build your casa. You keep the keys.
Connect Claude, or any app that speaks MCP, to your Mini Casa workspace. Ask it to make a casa, write the posts, connect your domain and check the DNS. It works through the same tools you do, with only the permissions you hand it.
106 tools in all. A new token starts read-only, and you add the rest one permission at a time.
Make me a casa for my pottery studio, add this week's kiln notes as a post, and connect clayandkiln.com.
sw_create_world { handle: "clayandkiln", template_id: "glaze-good" }→ draft casa madesw_create_post { kind: "note", title: "Kiln notes" }→ post addedsw_prepare_domain { domain: "clayandkiln.com" }→ records ready to addsw_verify_domain→ waiting for DNS
The whole casa, from first draft to live domain
These are some of the 106 tools your agent can call. It only sees the ones your token's permissions allow.
-
Make and publish casas
It picks a casita, writes the words and links, shows you a preview, publishes, and rolls back if you change your mind.
sw_list_templatessw_get_templatesw_create_worldsw_update_worldsw_get_worldsw_preview_revisionsw_publish_worldsw_rollback_world
-
Write posts
It adds notes, links, images and projects to a casa, edits them later, and sets how the posts section looks.
sw_create_postsw_update_postsw_list_postssw_update_posts_settingssw_delete_post
-
Connect a domain
It prepares your domain, hands you the exact DNS records to add, and checks them until the certificate is in place.
sw_prepare_domainsw_prepare_dns_plansw_get_dns_plansw_verify_domainsw_get_domainsw_list_domains
-
Keep track of your names
It brings every domain you own onto your shelf and tells you which ones renew soon and what they will cost.
sw_import_shelf_domainssw_list_shelf_domainssw_list_expiring_domainssw_get_shelf_summarysw_get_renewal_spend
-
Bring your links
It reads a link page you already have, shows you what it found, and adds only the links you choose.
sw_extract_link_pagesw_preview_profile_importsw_apply_profile_import
-
Readers and messages
It sets up the signup form, reads who subscribed and what visitors wrote, drafts replies, and pulls your numbers.
sw_update_audience_settingssw_list_subscriptionssw_list_inquiriessw_reply_inquirysw_query_analytics
-
Many domains at once
It turns a list of domains into casas, with a dry run first. Nothing runs until you approve the exact plan.
sw_create_portfoliosw_prepare_bulk_plansw_request_approvalsw_execute_bulk_plansw_get_bulk_runsw_pause_bulk_runsw_rollback_bulk_run
Two steps, a few minutes
-
Create a token
Sign in and open Developers in your dashboard. Choose New connection, pick MCP, tick the permissions it needs and copy the token. It is shown once, and lasts up to 90 days.
Open Developers -
Add it to your assistant
Paste one of these with your token in place of sw_mcp_... and keep the token in your app's settings, not in a chat.
Claude Code
Run this in a terminal. Add --scope user to use it from every project.
claude mcp add --transport http mini-casa https://mini.casa/mcp \
--header "Authorization: Bearer sw_mcp_..."
Claude Desktop
Add this to claude_desktop_config.json and restart Desktop. It uses the mcp-remote bridge, so you need Node.js.
{
"mcpServers": {
"mini-casa": {
"command": "npx",
"args": [
"-y",
"mcp-remote@0.15.2",
"https://mini.casa/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer sw_mcp_..."
}
}
}
}
Any other MCP client
Clients that accept an HTTP server with headers need only the address and the Authorization header.
{
"mcpServers": {
"mini-casa": {
"url": "https://mini.casa/mcp",
"headers": {
"Authorization": "Bearer sw_mcp_..."
}
}
}
}
Custom connectors in claude.ai and Claude Desktop's connector screen sign in with OAuth, which Mini Casa doesn't offer yet. Until it does, use Claude Code or the bridge above.
- Address
https://mini.casa/mcp- Transport
- Streamable HTTP, protocol 2026-07-28 and the 2025-11-25 handshake current clients use
- Sign-in
- Authorization: Bearer and your token
What a token can reach, and what it can't
- A token belongs to one workspace. Ask it about any other and it answers as if that workspace doesn't exist.
- It never does more than your own role allows, and your role is checked again on every call.
- You pick the permissions. Sensitive ones, like subscriber addresses or billing, are never ticked for you and ask you to sign in again first.
- Tokens expire after 30 days unless you choose otherwise, 90 at most.
- Revoke a token on the Developers page and it stops on the next request, including work it had queued.
- Text written by visitors, subscribers or imported pages reaches your agent marked as untrusted data, never as instructions.
Things an agent cannot do
- Approve its own plans. It can ask; only an owner can say yes, on the review page.
- Spend money. Checkout gives back a link that a person completes.
- Send mail to a person. Replies to visitors are saved as drafts for you to send.
- Delete or restore a casa or an account. That needs you, freshly signed in.
- Write DNS records at your provider. That is switched off for now: your agent prepares the records and checks them, and you add them where your DNS lives.
- Reach another workspace, or another person's casas.
Which assistants work with it?
Anything that speaks the Model Context Protocol over HTTP and can send an Authorization header. Claude Code works directly, Claude Desktop through the mcp-remote bridge, and there is a REST API for your own scripts.
Does connecting an agent cost extra?
No. Tokens come with every workspace, the free plan included; an owner or admin creates them. An agent gets the same plan limits you have, so anything your plan doesn't include stays out of its reach too.
Which permissions should I give it?
Start with the read-only defaults and add what the job needs: casa editing to build pages, domain permissions to connect a domain. Tools it has no permission for don't show up for it at all.
What if a token leaks?
Revoke it on the Developers page and it stops working on the next request. Only a hash is stored on our side, so nobody can read the token back, and every token expires within 90 days anyway.
Can my assistant report a bug to you?
Yes. Ask it to send feedback and it uses sw_submit_feedback, which any token can call. The report lands in our support inbox with your account, so we can follow up if you said that's fine. It asks you before including anything personal.
Is there an API for my own code?
Yes. The same operations are a REST API at /api/v1, described in an OpenAPI document. Pick REST API instead of MCP when you create the token.
Ready for a place of your own?
Pick a scene, add your words and links, and publish at a free Mini Casa address.