Use it when you use claude.ai, the Claude desktop or mobile apps, ChatGPT, or any other client that speaks MCP. It's also a tidy option for Claude Code.

How it works

This server is also an MCP server at https://app.slipdock.us/mcp. A client that connects gets a small set of tools that each do a whole step of the work: reading boards, cards, the guide and wiki pages; searching and asking; creating, updating, moving, commenting on, completing, archiving and deleting cards; making a card repeat (update_card's recurrence) or sleep until a day (its snoozed_until); leaving a card's handoff (leave_handoff: where it was left, which get_card shows first, or delete: true to delete one left by mistake); opening and answering a wait on a person (update_card's wait and answer_wait: several questions go as questions, one each, answered one by one with answers, or skipped with skip; get_card's waits: true lists the ones asked before); uploading and downloading the files on cards and pages (with a shell, link: true on upload_file or download_file gives a single-use URL, good for ten minutes, to curl, so neither the bytes nor a long-lived token pass through the model); lists, tags and boards; writing, sharing and reverting pages; listing, adding, changing and deleting a board's automations, for its owner; finding and installing Tools (list_tools, install_tool); the board's tidy-up suggestions (garden, and garden_decide for the ones the person approves; a proposed handbook edit is decided only on the board's Tidy-up page, signed in, and garden_decide and the CLI are refused with its link); a board's pulse and whether it is paused (get_board); and taking runner jobs. The tools go through the same permission checks as the API, so a connection can't do anything its token couldn't.

Read-only tools are marked as such, so a client can run them without asking each time. Deletes are marked destructive and need a confirmation in the call.

Answers are kept small, since a model pays for every one: each is compact JSON, sent once. get_board and list_cards give a brief of each card, without its description or comments (get_card has those; descriptions: true adds each description), and leave out or trim the done lists unless asked with include_done.

From claude.ai and the Claude apps

  1. In claude.ai, open Settings → Connectors → Add custom connector.
  2. Give it a name, and this address:
https://app.slipdock.us/mcp

Press Connect. You're sent here to sign in and approve the connection, or lower it to read-only first. A connector added once also works in the desktop and mobile apps. Each connection appears as a connected app under Account → API tokens, and deleting it there disconnects it.

claude.ai connects from Anthropic's servers, not from your browser, so this server has to be reachable from the internet over https.

From Claude Code

Add the server, then run /mcp in a session to sign in through the browser:

claude mcp add --transport http slipdock https://app.slipdock.us/mcp

Or pass a token from Account → API tokens instead of signing in:

claude mcp add --transport http slipdock https://app.slipdock.us/mcp --header "Authorization: Bearer <token>"

Claude Code connects from your own machine, so a server on your LAN or tailnet works as long as that machine can reach it.

Ready-made workflows

The server also offers prompts: workflows your client lists in its own menu, already holding the live board data they need. In Claude Desktop and claude.ai they are under the + menu; in Claude Code they are slash commands such as /mcp__slipdock__plan-my-day.

  • plan-my-day: my overdue, due-today and in-progress cards, and a plan for today.
  • weekly-review: the last 7 days, what slipped and what is stale.
  • triage: a board's Backlog or Inbox, card by card.
  • resume and leave-handoff: pick a card up where it was left, or leave it for the next person.
  • tidy-board: walk the gardener's tidy-up suggestions.

A connector remembers the list it was given, so after an upgrade that adds one, disconnect and connect it again. ChatGPT's connectors use tools only, so prompts don't show there.

Attach a board, card or page

You can also attach something to a conversation yourself, so the model starts with it: a board, a card, a wiki page, a saved view or My work. In Claude Desktop and claude.ai they are under the + menu; in Claude Code mention one, as in @slipdock:slipdock://card/42. The menu lists My work, your favourites and the pages pinned to your favourite boards, so favourite whatever you want to attach.

From ChatGPT and other clients

Any client that supports MCP's OAuth sign-in connects with just the address. Clients that take a URL and a header can use an API token instead, as above. In ChatGPT, add it as a connector, then upload the ChatGPT versions of the skills so it knows which tool to use for what.

What it can't do

  • Upload a file over 10 MB, such as most recordings, or transcribe one. Use the CLI or the API.
  • Set up runners, or run an automation now and read its callbacks. Do that on the board, or with the CLI.
  • Mirror your wiki to a folder on disk. Only the CLI does that.