Pi extension that authenticates with Cursor via PKCE OAuth and runs a local gRPC proxy translating OpenAI /v1/chat/completions to Cursor's protobuf/HTTP2 Connect protocol. Version 0.1.11, active development. Key differentiators: model deduplication by effort level, session management for multi-turn conversations, tool routing via MCP, and an architecture that lets Pi use Cursor's models without a direct Cursor IDE dependency.
npm install pi-cursor-providerNo compatibility data collected yet for this library.
Verified import paths — ran on the pinned version, not inferred.
Install the pi-cursor-provider extension, authenticate with Cursor via OAuth, select a model, and optionally configure reasoning effort. Shows CLI commands for setup.
Set PI_CURSOR_RAW_MODELS=1 environment variable to disable dedup and see all raw Cursor model IDs.
Ensure these peer dependencies are installed in the Pi environment. Run 'pi install' which resolves them.
Re-run /login cursor after restarting Pi. Tokens are stored in local state but may be invalidated.
Check the model list output by /model to see which models support reasoning effort. Use raw mode if effort mapping is unwanted.
Ensure your network allows HTTP/2 connections. Consider using a VPN if behind a restrictive firewall.
Install via pi: pi install npm:pi-cursor-provider, or manually add @mariozechner/pi-ai and @mariozechner/pi-coding-agent as dependencies.
Use import syntax or ensure your project is ESM (type: 'module' in package.json).
Convert to dynamic import: const { CursorProvider } = await import('pi-cursor-provider');Kill the existing process or set a custom port via environment variable (e.g., PI_CURSOR_PROXY_PORT=8080).
Re-run /login cursor to authenticate again.