Skip to content

README.md

Latest commit

 

History

History
64 lines (40 loc) · 4.11 KB

README.md

File metadata and controls

64 lines (40 loc) · 4.11 KB
Raw

vardoger — Cursor Plugin

Exposes vardoger as an MCP server that Cursor's agent can invoke to personalize your assistant.

Prerequisites

  • Cursor — cursor.com
  • One of:
    • uv (recommended, used by the shipped mcp.json) — vardoger is fetched on demand via uvx vardoger mcp.
    • pipx — for installs that pre-stage the CLI (see the fallback below).
  • Python 3.11+ — both uv and pipx will fetch a compatible Python if one is not already installed; see the main README prerequisites for manual installs.

Install from Cursor Marketplace

Install vardoger from the Cursor Marketplace and restart Cursor. The plugin ships a .cursor-plugin/plugin.json manifest plus an mcp.json that launches the server via uvx — no extra setup is needed if uv is on your PATH.

Install via pipx (fallback)

If uv is not available, run the classic setup once:

pipx install vardoger
vardoger setup cursor

This registers the vardoger MCP server in ~/.cursor/mcp.json using the pipx-resolved Python interpreter. Restart Cursor to activate.

Local Development Install

Point Cursor at this directory as a local plugin:

ln -s "$(pwd)/plugins/cursor" ~/.cursor/plugins/local/vardoger

Reload Cursor (Developer: Reload Window). The plugin's mcp.json runs uvx vardoger mcp; if you want to test against an in-tree build, swap the command to your local .venv/bin/vardoger mcp while iterating.

Usage

Ask the Cursor agent:

  • "Personalize my assistant"
  • "Run vardoger"
  • "Analyze my conversation history"

The agent will call the vardoger_personalize tool, which returns step-by-step orchestration instructions. The agent then follows them automatically — preparing batches, summarizing, synthesizing, and delivering the result.

Where the personalization lands

vardoger analyses your global Cursor conversation history, so the output it produces is user-level (applies to every workspace), not project-level. Delivery defaults reflect that:

  • Default — User Rules (copy-paste): vardoger_write saves the rendered block to a convenience copy-source file at ~/.vardoger/cursor-user-rules.md and prints that absolute path at the top of its response. Cmd+click the path in Cursor's chat to open the file in an editor tab, then copy the whole file into Cursor Settings → Rules → User Rules. Pasting there is what actually activates the rules — Cursor's User Rules live in its settings database, not on disk, so the copy-source file is just there so you don't have to dig for the block inside a collapsed tool-call card. Re-running vardoger_write overwrites that file with the latest generation; rejecting via vardoger_feedback reject either rewrites it with the previous generation or deletes it. Edit the block freely once pasted — the bullets are starting points derived from patterns in your chat history, not commandments.
  • Opt-in — project-scoped file: Ask the agent "also drop this into my current workspace" and it will call vardoger_write with project_path=<your workspace root>. vardoger writes <project>/.cursor/rules/vardoger.md only if that directory (or one of its ancestors) looks like a real project (contains .git, a language manifest, AGENTS.md, or an existing .cursor/). If it doesn't, the write is refused with an actionable error — vardoger will not silently drop a rules file into $HOME or any other non-project location.

Reusing a personalization from another workspace

If you've already curated a vardoger.md in another Cursor workspace, tell the agent: "you can also check workspace X and workspace Y." The agent calls vardoger_import with those paths; vardoger returns any vardoger.md it finds so the agent can offer to reuse, merge, or ignore it before running a fresh analysis.

Uninstall

Uninstall from the Cursor Marketplace panel (or remove the "vardoger" key from ~/.cursor/mcp.json if installed via vardoger setup cursor).