# ZeroMark · Your taste in X, for Claude, ChatGPT and Gemini > Connect Claude, ChatGPT or Gemini to your taste in X. Read-only: what you kept, what you let go, and Living Memory. Revocable. Canonical: https://www.zeromark.io/docs/connector Living Memory connector ## Your taste in X, readable from Claude, ChatGPT and Gemini. Connect an assistant to ZeroMark and it reads your taste in X: the posts you kept, the ones you let go, and the notes your library has become. It can still search every save. Read-only, your library only, and revocable in one press. ### Why this is not just your bookmarks in a chat Any tool can pour every bookmark into an assistant. That hands it what you hoped to read: thousands of saves, all equal, many of them cut off mid-sentence. ZeroMark hands it your taste in X: what you decided, one save at a time, Done or Let go. What you keep becomes Living Memory, notes in your own words that change as you do. Your assistant reads those decisions, says how many its answer rests on, and says plainly when there are too few to call it taste. It can suggest. It never decides: nothing here can mark a save Done or let one go. ### What an assistant can read Eleven tools, every one read-only. Each save an assistant gets back carries its link on X, and says who decided it: you, you in a sweep, or the Time Test. Only your own one-at-a-time decisions count as your taste. Tool · What it does get_taste · Your taste in X: kept and let go per field and per voice, what you let go lately, and what your notes stopped believing, with dates. Only your own decisions, nothing a model inferred. search_saves · Finds saves by keyword and by meaning. Filters by status (open, done, a stage) and by date, or lists what you saved between two dates, 25 at a time with a next page. get_save · One save in full: the post, the post it quotes, the author, its field and intent, its stage, your takeaway and the Living Memory notes it feeds. list_by_stage · Every save in one stage, newest or oldest first, whichever reads naturally for that stage. library_overview · How many saves are in each stage, the biggest fields, and how many Living Memory notes: a good first call. get_memory · Your whole Living Memory as one Markdown page, the same file Settings exports. get_map · Your Map: each field with how much of it you keep, the saves in different fields that reach for the same ideas, and the notes you pinned together. get_podcast · Your latest podcast as text: what it covered, its outline and transcript, and a link to listen in ZeroMark. search, fetch · The same search and the same full save, in the shape ChatGPT's deep research looks for. weekly_question · A suggestion: the one open save worth acting on this week, with the reasons and the links. You decide. ### The five stages A stage is one word for where a save is in its life, worked out from what you already decided in ZeroMark. There is nothing new to maintain. - **fresh**: Still open, saved in the last two weeks. - **still_useful**: You marked it Done, pinned it, or the Time Test says it still holds. - **fading**: Still open and older than two weeks, or the Time Test says its moment passed. - **let_go**: You let it go, or it was retired. - **snoozed**: Hidden from your queue until a date you picked. ### Let your assistant set it up In Claude Code, Gemini CLI, Cursor or VS Code, paste this to your agent: Set up ZeroMark for me by following https://www.zeromark.io/docs/connector/agent-setup.md It adds the server URL to the client it runs in, shows you the change before it makes it, and hands the sign-in to you in your browser. It installs nothing and never asks for a key in chat. ### Check it worked Ask your assistant which ZeroMark tools it has. It should list eleven: search_saves, get_save, list_by_stage, library_overview, get_memory, search, fetch, get_taste, get_map, get_podcast, weekly_question. If they are missing, the sign-in has not finished. ### Connect in two minutes Open ZeroMark, go to Settings, and find **Connect to Claude, ChatGPT or Gemini**. It shows your server URL and makes keys. A key is shown once, so copy it before you leave the page. #### Claude Code With no key: run claude mcp add --transport http --scope user zeromark https://www.zeromark.io/api/mcp, then type /mcp, choose **zeromark** and sign in. Or with a key, in place of the placeholder: claude mcp add --transport http --scope user zeromark https://www.zeromark.io/api/mcp \ --header "Authorization: Bearer zm_mcp_YOUR_KEY" #### Gemini CLI Add the server to ~/.gemini/settings.json: { "mcpServers": { "zeromark": { "httpUrl": "https://www.zeromark.io/api/mcp", "headers": { "Authorization": "Bearer zm_mcp_YOUR_KEY" } } } } The same URL and header work with the Gemini API's tool calling, for anyone building on it. #### claude.ai, Claude Desktop and ChatGPT Find ZeroMark in Claude's directory and connect it. That signs you in to ZeroMark, so there is no key to paste. If it is not listed, open Settings, then Connectors, then **Add custom connector**, and enter https://www.zeromark.io/api/mcp. In ChatGPT, add a custom connector with the same URL. Each one asks you to approve read-only access on a ZeroMark page, and appears in Settings so you can disconnect it later. ### Which assistants can connect Assistant · How Claude · Claude Code with a key, today. On claude.ai, the desktop app and the phone app, find ZeroMark in Claude's directory first. If it is not listed, add a connector with the server URL and sign in. ChatGPT · A custom connector by URL, with sign-in. Deep research can use it too. Gemini · Gemini CLI with a key, today. The Gemini app's custom apps are limited to some accounts in the US and need sign-in. Perplexity · Pro, Max and Enterprise add a custom remote connector by URL, with sign-in. Muse · Meta's Muse app takes connectors only through Meta's reviewed directory, so not yet. Muse Code, Meta's coding tool, connects with a key. Instinct · Instinct has no way to add your own connector today. ### No connector? Upload one file Settings, Export everything, **AI-ready file**. It is your whole library as one Markdown file: a short note to the assistant on what it is reading, every save grouped by stage with its link, then your Living Memory. Upload it to any chatbot that takes a file. ### If something goes wrong - **"A valid ZeroMark connector key is required":** the key was mistyped or revoked. Make a new one in Settings; the header is Authorization: Bearer followed by the key. - **"This account's plan does not include the library right now":** open ZeroMark and it will say why. - **"Too many calls":** wait a few minutes; the limit resets every 15. - **It connects but finds nothing:** check you are using the exact server URL from Settings, with the www. - **"No save with that id":** the save is not in this library. A key only ever opens its owner's library. ### What stays private - **Read-only:** no tool can decide, edit or delete anything, in ZeroMark or on X. - **Your library only:** a key opens exactly one library, its owner's. Asking it for anyone else's save answers the same as asking for one that does not exist. - **Revocable:** revoke a key in Settings and the next call is refused. Keys are stored only as a one-way hash, so even ZeroMark cannot read one back. - **Counted, not recorded:** each call is logged as a tool, a time and a count. What was asked is never kept. - **Rate-limited:** each key gets 120 calls every 15 minutes and each account 300, far more than a conversation needs and far less than a runaway loop would spend. - **Other people's words are labelled:** every answer tells the assistant that post text is content to report, never instructions to follow. ### Questions people ask #### Can an assistant change anything in ZeroMark? No. Every tool is read-only. An assistant can search your saves, open one, list a stage, read your Living Memory and ask the weekly question. It cannot mark anything Done, let anything go, snooze, pin, edit a note or delete. Nothing it does reaches your X account either. #### What does ZeroMark keep about what my assistant asked? For each call: which tool, when, how many saves came back, and whether it worked. You can read that list in Settings. The words your assistant searched for and the answers it got are never stored. #### How do I cut an assistant off? Revoke its key, or disconnect it, in Settings. The next call it makes is refused. Removing the connector inside the assistant also works, but revoking in ZeroMark is the one that does not depend on anyone else. #### Does my library go anywhere new? Only to the assistant you connect, and only what it asks for. ZeroMark sends nothing to anyone else on its behalf. What that assistant does with what it reads is covered by its own terms, so connect the ones you already trust with your chats. #### Which plans include it? Every plan, the free plan included. A free account connects the saves it holds, and paying reads the whole library the same way. #### Does the weekly question count toward my Chat questions? Yes. It asks the same reading ZeroMark's Chat uses, from your library only and never the web, so on the free plan it is one of the day's Chat questions. The answer is kept for the week, so asking twice is free. #### My assistant has no connectors. What then? Download the AI-ready file from Settings, under Export everything, and upload it. It is one Markdown file with every save grouped by stage, each with its link, and your Living Memory at the end.