Let your editor's agent edit live documents (MCP)

Cursor, Windsurf, VS Code, Claude Code, Codex and other MCP clients can edit your wickrunAI live documents through wickrunAI's live-document tools. You keep your editor: others see the changes as they happen on the web, phone and desktop, the edits are signed with a name such as "your Cursor", and every step stays in the version history, where it can be compared and restored.

Before you start

With the desktop app (recommended): wickrunAI desktop 6.1 or later, running and signed in. Documents in encrypted spaces can be edited too.
Without the desktop app: use the cloud connector, with a connector address you create in wickrunAI on the web. Documents in encrypted spaces cannot be edited this way.

Connect through the desktop app

In the desktop app open Settings → Tools → Live documents and click "Copy editor MCP config". You get a complete mcpServers block with paths on this computer and no password or key.
Paste it into your client:

Cursor: .cursor/mcp.json in the project, or ~/.cursor/mcp.json for every project.

{
  "mcpServers": {
    "wickrun-docs": {
      "command": "the command copied from wickrunAI",
      "args": ["the arguments copied from wickrunAI"]
    }
  }
}

Claude Code: .mcp.json at the project root (same format), or add it to your own user configuration:

claude mcp add --scope user wickrun-docs -- <the command and arguments copied from wickrunAI>

Codex: ~/.codex/config.toml

[mcp_servers.wickrun-docs]
command = "the command copied from wickrunAI"
args = ["the arguments copied from wickrunAI"]

Windsurf, VS Code and other MCP clients: add the same mcpServers block in their MCP settings.

Restart the client or refresh its MCP list; wickrun-docs should appear.

The paths differ from computer to computer. In a repository your team shares, put the block in each person's user configuration rather than committing it.

Connect through the cloud connector

Create a connector address in wickrunAI on the web (it looks like https://wickrunai.com/mcp/<your connector token>) and add it in your client as a remote MCP server:

{ "mcpServers": { "wickrun-docs": { "url": "https://wickrunai.com/mcp/<your connector token>" } } }

In Claude Code:

claude mcp add --scope user --transport http wickrun-docs https://wickrunai.com/mcp/<your connector token>

The address works like a password for your account's connector: keep it in your own user configuration, never commit it and never send it to anyone. Creating a new address makes the old one stop working at once.

What the agent can do

Tool	What it does
`wickrun_doc_list`	Lists the live documents you can open; without a location, the ones you changed recently
`wickrun_doc_open`	Opens a document; the agent joins it as a member such as "your Cursor"
`wickrun_doc_apply`	Changes the document: replace the whole text, or find one passage and replace it
`wickrun_doc_versions`	Shows the version history
`wickrun_doc_version_text`	Reads one version
`wickrun_doc_mark_version`	Saves a Work version after a piece of work
`wickrun_doc_presence`	Shows who is in the document right now

What the edits look like

The document's member bar shows "R's Cursor" (R is your name) with a robot badge, apart from you.
The agent changes only the text it changed. What others write at the same time is merged in and never overwritten.
The version history shows which changes came from the agent; compare them, or restore an earlier version.

Limits you may run into

The agent works with your account and your permissions; it has no account of its own. Documents you can only view, it can only read.
Through the desktop app, wickrunAI must be running and signed in. An idle agent leaves the document after a while; open it again.
Documents in encrypted spaces can be edited only through the desktop app; the cloud connector answers "open it on your computer".
A file nobody has opened with "Edit together" yet has an empty live document. Open it once in wickrunAI first ("Edit together" in Drive).
If someone changed the document after the agent read it, replacing the whole text may be refused. Have the agent open the document again, or use find and replace.