Connect your agent to Antescript
Claude Code, Claude Desktop, Cursor or any client that speaks MCP can read and write your pages as markdown. It works as you: it reaches the pages you can reach and changes only what you could change.
The MCP server
Antescript’s MCP server is at https://mcp.antescript.dev/mcp, hosted on MCPCloud. It gives an agent the API’s calls as tools. Connecting takes two steps:
- Open the connect page (opens in a new tab) and continue with Antescript. Sign in, name the agent and choose Read, or Read and write. There’s no token to copy or paste.
- Add the server to Claude Code, Claude Desktop, Cursor or any MCP client with the snippet the connect page gives for it, and sign in when the client asks.
An agent per app
Connect Claude Code, claude.ai and Cursor with the same sign-in and each works as an agent of its own: every Claude app is Claude, Cursor is Cursor. People on a page see which one wrote what, and its history keeps the mark it had. Under Agents you can stop one app alone and allow it again; Disconnect stops them all.
Your agent never holds the token
Signing in gives MCPCloud an hour’s access, renewed until you disconnect; it adds it to each call on its way to Antescript, and the agent holds only a short-lived pass of its own. MCPCloud learns your account’s id, name and email address to know who you are. The connection is listed under Agents; Disconnect ends it at once.
What an agent can do
The calls below, by their names in the API’s contract. Every write lands in the page the way typing does, so people editing it at the same time keep their text; a write under one heading leaves the rest of the page alone. A comment an agent writes shows under its token’s name with the mark, and typing @ in a comment and picking your agent asks it, which it finds by checking its mentions.
me- Whose token it is and what it may do.
listWorkspaces- The workspaces you belong to, with your role in each.
listPages- Every page in a workspace, with where it sits in the tree; or only the pages edited since a time.
searchPages- Titles and text in a workspace that match, with a snippet.
getPage- A page, its content as markdown, and its sections by heading.
getPublishedPage- The page behind a published address, as markdown, when you may read it.
createPage- A new page, with content if you give it, under a parent or at the top. Needs the write scope.
renamePage- A page’s new title. Needs the write scope.
replacePageMarkdown- A page’s content, replaced with markdown. Needs the write scope.
appendPageMarkdown- Markdown added at the end of a page. Needs the write scope.
replaceSection- What is under one heading, replaced; the heading and the rest of the page stay. Needs the write scope.
appendToSection- Markdown added at the end of one heading’s section. Needs the write scope.
pageChanges- What changed on a page since a time, as a diff of its markdown.
pageActivity- Who changed a page and when: people’s edits, agents’ writes, renames, moves, publishing and threads.
listThreads- A page’s comment threads, each comment as text, and the text each thread is about.
createThread- A comment thread, placed beside a quoted line of the page or on the page as a whole. Needs the write scope.
replyToThread- A reply in a thread, under the token’s name. Needs the write scope.
setThreadResolved- A thread resolved, or reopened. Needs the write scope.
listMentions- The comments that mention this agent by name, or you: what you were asked, to answer in the thread. Each app’s agent sees its own.
createDesign- A new design: a canvas people keep editing, SVG, Mermaid or an HTML page, and optionally a figure of it on a page. Needs the write scope.
getDesign- A design’s source, format and colours, and where it renders.
replaceDesign- A design’s source replaced, or its title or colours. Needs the write scope.
editDesignElements- Elements added, changed or removed on a canvas design, by id, in one edit. Needs the write scope.
placeDesign- A design shown as a figure at the end of a page. Needs the write scope.
startHtmlEmbed- An upload address for an HTML file, such as a Claude artifact. Needs the write scope.
placeHtmlEmbed- The uploaded HTML file on the page, sandboxed, as it looks. Needs the write scope.
listGroups- A workspace’s sidebar groups: the shared ones and your private ones.
createGroup- A new sidebar group, shared unless you make it private. Needs the write scope.
updateGroup- A group’s new name or icon. Needs the write scope.
movePage- A page and its subpages into a group, under another page, or to the top. Needs the write scope.
setPageIcon- A page’s icon: one emoji, or none. Needs the write scope.
trashPage- A page and its subpages into the trash, where anyone who edits can restore them for 30 days. Needs the write scope.
restorePage- A trashed page back where it was. Needs the write scope.
listTrash- What is in a workspace’s trash, newest first.
Pages come and go in Antescript’s markdown dialect, so callouts, toggles and checklists survive the trip.
Things to ask it
Plain requests work; the agent picks the tools. A few that use what Antescript is for:
- “Read the PRD in Antescript and turn its open questions into a checklist at the end.”
- “Rewrite the Rollout section of the spec so it matches what we just shipped; leave the rest alone.”
- “Draw the checkout flow from that page as a diagram and put it under the Flow heading.”
- “What changed on the launch plan since Monday, and who changed it?”
- “Check your mentions and answer what you were asked in each thread.”
- “File the drafts from this week into the Launch group and give each an icon.”
What it can’t do
- Delete for good, share or publish. Those stay with people. An agent may move a page to the trash, but never empty it, so a person can always undo what it did, and its trash and moves show in the page’s activity under its name.
- Change who sees a page. A move that would make a page private, or shared again, is refused; so is changing a group’s scope or deleting a group.
- Reach past you. A token acts as the person who made it: it sees the pages they can open and writes only where they could edit.
- Go where a workspace keeps it out. The workspace’s owner chooses under Agents, in its settings, whether agents may read and write there, only read, or nothing.
- Write with a read token. Creating, renaming, arranging and changing content need the write scope.
Scopes, rates and daily calls
- Scopes
readlists, searches and reads;writealso creates, renames and changes content. You choose when you make the token.- Rate
- 120 calls a minute per token, counted per clock minute. Past it the API answers
429, withRetry-Aftergiving the seconds until the minute ends. - Calls a day
- Counted against the plan of the account that owns the workspace, starting again at 00:00 UTC: 100 on Free, 1,000 on Starter and 1,000 per editor on Team. Past it the API answers
402withplan_limit. - Size
- Request bodies up to 1 MB.
- Record
- Each call is logged with its token, the call and the page, and kept for 30 days; every response carries an
X-Request-Id.
The REST API
Anything that can make an HTTP request works without MCP. The API is JSON at https://api.antescript.dev/v1, with your token as Authorization: Bearer. Its OpenAPI contract, which the MCP server’s tools are made from, is at https://api.antescript.dev/v1/openapi.json.
curl -s https://api.antescript.dev/v1/me \ -H "Authorization: Bearer $ANTESCRIPT_TOKEN"
An error names its code and the request, and the status says what kind it is: 400 for a body or a time the call doesn’t take, 401 for no token or an unknown one, 403 for a scope or role that doesn’t allow the call, 404 for what the token can’t reach, 413 for too large, 429 for too fast and 502 when the page’s room didn’t answer.
{
"error": {
"code": "insufficient_scope",
"message": "This token lacks the write scope."
}
}Privacy and support
An agent sees only what you could see and what it asks for: the pages, comments and designs its calls return. Antescript keeps no copy of your conversations with it and sends nothing to an AI provider; the agent is yours, and so is the model behind it. Each call is logged with the token, the call and the page, and the log is kept 30 days. The privacy policy has the rest. Questions, or something an agent did that it shouldn’t have, go to support@antescript.app.
The guide’s Connect your agents page has the same steps as the app shows them, and a few things to ask.