
lvndr-agent
io.github.toolfora/lvndr-agent
Shopify store change log with attribution and precise undo, restore points and agent runs.
Documentation
LVNDR agent kit
Everything an AI agent needs to change a Shopify store safely with
LVNDR by Toolfora, the store time machine for Shopify:
the MCP server connection, the lvndr skill, plugins for Claude Code, Cursor and Codex, a Cursor
rule, the MCP registry entry and the zero-dependency lvndr CLI.
LVNDR records changes to the parts of a store it tracks (products, variants, inventory,
collections, pages, blogs and posts, media, publication, metafields and definitions, menus,
themes), says who made each one as far as Shopify reveals it, and puts it back with a preview
first. Every theme file is tracked and put back through the merchant's own Shopify CLI (lvndr theme restore); orders and customers are not tracked. An agent starts a run with a restore point before it
writes; afterwards it lists exactly what the run changed and undoes the whole run, one batch or one
field. The LVNDR Shopify app must be installed on the store. This repository holds no app source.
- Developer page: https://toolfora.com/lvndr/developers
- For language models: https://toolfora.com/lvndr/llms.txt
- OpenAPI: https://lvndr.toolfora.com/api/v1/openapi.json
- Operator guide:
docs/AGENT-PLAYBOOK.md
Step 0: install LVNDR on the store
Install the LVNDR Shopify app on the store and wait for its first snapshot (the app's change log
shows the progress; a few minutes for most catalogues). Then connect your agent and call
lvndr_status (CLI: lvndr status) to confirm which store it reached.
Connect
MCP endpoint: https://lvndr.toolfora.com/mcp (Streamable HTTP).
The first option for every client is keyless: give it only the URL, and it signs in through the
Shopify admin with OAuth (you pick the store and approve the scopes). The second option is an API
key: create one in LVNDR > Settings > Agents & API (starts with lv_, shown once), export
it, and send it as Authorization: Bearer <key>:
export LVNDR_API_KEY=lv_...
API access is included on every LVNDR plan and free on development stores. New keys can read and undo; rollback and export are opt-in per key.
Claude Code
Plugin (MCP server + skill; signs in with OAuth on first use):
/plugin marketplace add toolfora/lvndr-agent
/plugin install lvndr@lvndr
Or the MCP server alone, keyless (then run /mcp in Claude Code to sign in):
claude mcp add -t http lvndr https://lvndr.toolfora.com/mcp
With an API key instead:
claude mcp add -t http -H "Authorization: Bearer $LVNDR_API_KEY" lvndr https://lvndr.toolfora.com/mcp
Cursor
Keyless: add the URL to ~/.cursor/mcp.json (or .cursor/mcp.json in a project); Cursor signs in
with OAuth when the server is first used:
{ "mcpServers": { "lvndr": { "url": "https://lvndr.toolfora.com/mcp" } } }
With an API key, one-click install (Cursor reads LVNDR_API_KEY from its environment):
cursor://anysphere.cursor-deeplink/mcp/install?name=lvndr&config=eyJ1cmwiOiJodHRwczovL2x2bmRyLnRvb2xmb3JhLmNvbS9tY3AiLCJoZWFkZXJzIjp7IkF1dGhvcml6YXRpb24iOiJCZWFyZXIgJHtlbnY6TFZORFJfQVBJX0tFWX0ifX0=
The config value is the base64 of
{"url":"https://lvndr.toolfora.com/mcp","headers":{"Authorization":"Bearer ${env:LVNDR_API_KEY}"}}.
The Cursor plugin in .cursor-plugin/ bundles the URL-only server
(OAuth), the skill and the rule in rules/lvndr.mdc.
VS Code
vscode:mcp/install?%7B%22name%22%3A%22lvndr%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Flvndr.toolfora.com%2Fmcp%22%2C%22headers%22%3A%7B%22Authorization%22%3A%22Bearer%20%24%7Benv%3ALVNDR_API_KEY%7D%22%7D%7D
As a web link: Install in VS Code.
VS Code must be started from a shell where LVNDR_API_KEY is set.
Codex
In ~/.codex/config.toml:
[mcp_servers.lvndr]
url = "https://lvndr.toolfora.com/mcp"
bearer_token_env_var = "LVNDR_API_KEY"
.codex-plugin/plugin.json packages the skill and points at
.mcp.json, the URL-only server (codex plugin marketplace add toolfora/lvndr-agent).
If your Codex build does not do MCP OAuth, use the config.toml snippet above with a key.
Claude Desktop
Through mcp-remote, in
claude_desktop_config.json:
{
"mcpServers": {
"lvndr": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://lvndr.toolfora.com/mcp", "--header", "Authorization:${LVNDR_AUTH}"],
"env": { "LVNDR_AUTH": "Bearer lv_..." }
}
}
}
CLI
npm i -g @toolfora/lvndr # once: installs the `lvndr` command
lvndr login # paste the key; saved to ~/.config/lvndr/config.json
lvndr status
lvndr changes --since 24h
lvndr undo <changeId> # preview; prints the --plan-hash to write it
lvndr undo <changeId> --run --plan-hash <hash> # write exactly that plan
lvndr guard -- shopify store execute --allow-mutations ... # run + restore point around a command
Every command, flag and exit code: cli/README.md.
Skill only
The skill follows the Agent Skills format. Copy
skills/lvndr/ into your agent's skills directory (for Claude Code,
~/.claude/skills/lvndr/), or npx skills add toolfora/lvndr-agent. Endpoint and tool reference:
skills/lvndr/references/.
MCP registry
server.json describes the remote server for the official MCP registry as
io.github.toolfora/lvndr-agent.
Layout
| Path | What |
|---|---|
.claude-plugin/ | Claude Code plugin and marketplace manifests |
.cursor-plugin/ | Cursor plugin manifest and its MCP config |
.codex-plugin/ | Codex plugin manifest |
.mcp.json | The LVNDR HTTP MCP server, URL only (OAuth sign-in; no secrets) |
skills/lvndr/ | The safe-change skill and its references |
rules/lvndr.mdc | Cursor rule (opt-in) |
cli/ | The lvndr npm package |
server.json | MCP registry entry |
docs/AGENT-PLAYBOOK.md | Guide for the person running agents |
License
MIT, see LICENSE. The LVNDR app itself is a hosted service and is not part of this
repository.
No installation packages available.