Works with the AI assistant you already use.
PicPrep is built to work with an assistant. It edits alongside you in the app, every change marked as its own with its reason, and you keep or reject each one.
What it can do
- Open your projects and see them: your marks, notes, posts and slides, and what the window shows.
- Ask you a question or offer choices. They wait in PicPrep's Questions inbox, with a count on the Assistant button, until you answer there, with a note if you like. Nothing pops up.
- Lay out slides and posts: split a cell, stack two slides, put photos in a post. Each change appears at once, in its colour, with its reason, even while the window is in the background.
What stays yours
- Keep or reject every change, one by one or all at once, or edit on top of it. Comment on a change and the assistant reads it.
- Approving, exporting, settings and your licence. It cannot do them. Deleting, restoring a version and adding photos it can only offer, for you to apply.
- Pause it at any time from the Assistant button. While it looks at your window, PicPrep says so.
Private by design
The connector talks only to PicPrep on your own computer (127.0.0.1). PicPrep uploads nothing.
An assistant that runs in the cloud receives what it views, such as screen descriptions and pictures of the window, under its own terms, not PicPrep's. Pause viewing whenever there is something on screen it should not see.
Set it up
You need PicPrep installed, and Node.js 18 or newer for npx (Claude Desktop's one-click bundle needs neither). The connector, picprep-mcp, is free and has no dependencies; read tools work without a licence, and changes need a subscription or an active trial.
Any assistant that runs a local tool: command npx, arguments -y picprep-mcp.
Claude Code
Claude Code
Plugin (connector and skills together):
claude plugin marketplace add picprep/picprep-agents
claude plugin install picprep@picprep
Inside a session the same is /plugin marketplace add picprep/picprep-agents, then /plugin install picprep@picprep. The plugin runs npx -y picprep-mcp and adds the five PicPrep skills.
Connector only:
claude mcp add picprep -- npx -y picprep-mcp
On native Windows (not WSL), npx must go through cmd: claude mcp add picprep -- cmd /c npx -y picprep-mcp.
Check the connection with /mcp in a session, or npx -y picprep-mcp --check in a terminal.
Claude Desktop
Claude Desktop
One click: download picprep.mcpb from the latest release of this repository and open it; Claude Desktop asks to install the PicPrep extension. It carries picprep-mcp inside, so nothing else is needed.
By hand: Settings → Developer → Edit Config, then add to claude_desktop_config.json:
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}
Restart Claude Desktop. Skills: upload the folders under skills/ in Settings → Capabilities → Skills, as zip files.
OpenAI Codex
OpenAI Codex (CLI, app and IDE extension)
Plugin (connector and skills together):
codex plugin marketplace add picprep/picprep-agents
then install PicPrep from the plugin list (/plugins in the CLI). The plugin is .codex-plugin/plugin.json, with the connector in mcp.json and the skills in skills/.
Connector only: codex mcp add picprep -- npx -y picprep-mcp, or in ~/.codex/config.toml:
[mcp_servers.picprep]
command = "npx"
args = ["-y", "picprep-mcp"]
Skills without the plugin: copy the folders under skills/ into ~/.agents/skills/ (or a project's .agents/skills/).
Cursor
Cursor
One click: Add PicPrep to Cursor
(cursor://anysphere.cursor-deeplink/mcp/install?name=picprep&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsInBpY3ByZXAtbWNwIl19 - the config is {"command":"npx","args":["-y","picprep-mcp"]} in base64.)
By hand: ~/.cursor/mcp.json (every project) or .cursor/mcp.json (one project):
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}
Skills: copy the folders under skills/ into ~/.cursor/skills/ or ~/.agents/skills/ (or a project's .cursor/skills/ or .agents/skills/).
Google Antigravity
Google Antigravity
Add to ~/.gemini/config/mcp_config.json (every workspace) or .agents/mcp_config.json (one workspace), or through the agent panel's MCP servers → Manage → View raw config:
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}
Skills: copy the folders under skills/ into ~/.gemini/config/skills/ (every workspace) or .agents/skills/ (one workspace). The Antigravity CLI reads ~/.gemini/antigravity-cli/skills/ for every workspace.
Gemini CLI
Gemini CLI
Extension (connector and skills together):
gemini extensions install https://github.com/picprep/picprep-agents
The extension is gemini-extension.json at the root of this repository; it runs npx -y picprep-mcp and brings the skills in skills/.
Connector only: in ~/.gemini/settings.json:
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}VS Code (Copilot)
VS Code (GitHub Copilot agent mode)
.vscode/mcp.json in a workspace, or the user mcp.json (command palette: "MCP: Open User Configuration"). VS Code needs "type": "stdio", and its key is servers, not mcpServers:
{
"servers": {
"picprep": { "type": "stdio", "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}
Or: code --add-mcp '{"name":"picprep","type":"stdio","command":"npx","args":["-y","picprep-mcp"]}'. Skills: copy the folders under skills/ into a workspace's .github/skills/ or .agents/skills/.
Devin CLI
Devin CLI
Local only: a Devin cloud session runs on another machine and cannot reach the PicPrep app on yours. Use the Devin CLI on the computer where PicPrep is installed.
devin mcp add picprep -- npx -y picprep-mcp
Or in ~/.config/devin/mcp_config.json (every project) or .devin/mcp_config.json (one project):
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}OpenCode
OpenCode
In opencode.json (a project) or ~/.config/opencode/opencode.json (every project). OpenCode's key is mcp, and the command is one array:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"picprep": { "type": "local", "command": ["npx", "-y", "picprep-mcp"], "enabled": true }
}
}
Skills: copy the folders under skills/ into ~/.config/opencode/skills/ or ~/.agents/skills/ (or a project's .opencode/skills/ or .agents/skills/).
Windsurf
Windsurf (Devin Desktop)
Windsurf's Cascade reads ~/.codeium/windsurf/mcp_config.json; Devin Desktop, its successor, reads ~/.config/devin/mcp_config.json (%APPDATA%\devin\mcp_config.json on Windows). Either takes:
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}
Then refresh the MCP servers in the Cascade panel.
Zed
Zed
In Zed's settings.json (Zed → Settings → Open Settings). Zed calls MCP servers context servers:
{
"context_servers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"], "env": {} }
}
}Cline
Cline
Cline panel → MCP Servers → Installed → Configure MCP Servers, which opens cline_mcp_settings.json:
{
"mcpServers": {
"picprep": { "command": "npx", "args": ["-y", "picprep-mcp"], "disabled": false }
}
}Goose
Goose
For one session: goose session --with-extension "npx -y picprep-mcp".
To keep it: goose configure → Add Extension → Command-line Extension, name picprep, command npx -y picprep-mcp. That writes, in ~/.config/goose/config.yaml:
extensions:
picprep:
name: picprep
type: stdio
cmd: npx
args: [-y, picprep-mcp]
enabled: true
timeout: 300oh-my-pi
oh-my-pi (omp)
In ~/.omp/agent/mcp.json (every project) or .omp/mcp.json (one project):
{
"mcpServers": {
"picprep": { "type": "stdio", "command": "npx", "args": ["-y", "picprep-mcp"] }
}
}If it does not connect, run npx -y picprep-mcp --check: it says where it looks, whether PicPrep is running, and what it would start.