cursor-pi

cursor-pi registers a cursor-cli provider in Pi and delegates every model call to the local Cursor CLI with cursor-agent -p --mode ask --trust.

This extension is intentionally a CLI bridge. It does not use Cursor's HTTP API or Pi's built-in providers as a fallback. If cursor-agent is missing, not logged in, or fails, the Pi model turn fails with setup guidance instead of silently using another transport.

Requirements

• Pi Coding Agent
• Cursor CLI installed and available on the same machine:

cursor-agent --version

Install it if needed:

curl https://cursor.com/install | sh

• Logged in to your Cursor account:

cursor-agent login

Install

From this repository:

npm run install-extensions

Then restart Pi or run:

/reload

For one-off development testing without copying:

pi -e ./extensions/cursor-pi/src/index.ts

Usage

Pick provider cursor-cli from /model, or start Pi directly:

pi --provider cursor-cli --model auto

Bundled model aliases mirror common cursor-agent models ids:

| Pi provider | Pi model id | Passed to Cursor CLI | |-------------|-------------|----------------------| | cursor-cli | auto | cursor-agent -p --model auto --mode ask --trust | | cursor-cli | composer-2.5 | cursor-agent -p --model composer-2.5 --mode ask --trust | | cursor-cli | gpt-5.3-codex-high | cursor-agent -p --model gpt-5.3-codex-high --mode ask --trust | | cursor-cli | claude-sonnet-5-thinking-xhigh | cursor-agent -p --model claude-sonnet-5-thinking-xhigh --mode ask --trust | | cursor-cli | gemini-3.7-flash-high | cursor-agent -p --model gemini-3.7-flash-high --mode ask --trust |

Run cursor-agent models (or /cursor-pi models) to see every id available on your account; effort variants like *-fast, *-xhigh are separate model ids.

Print-mode smoke test:

pi -p --provider cursor-cli --model auto "Reply with exactly OK"

Direct Cursor CLI transport check:

echo "Reply with exactly OK" | cursor-agent -p --output-format text --mode ask

Commands:

/cursor-pi            — read-only TUI status/config menu (TUI only)
/cursor-pi status     — full status: binary version, auth, registered models
/cursor-pi verify     — installation + login verification with fix guidance
/cursor-pi usage      — current plan tier and account info from `cursor-agent about`
/cursor-pi models     — registered models plus account models from `cursor-agent models`
/cursor-pi test       — print smoke-test commands
/cursor-pi help

Read-only TUI menu

In the interactive TUI, bare /cursor-pi opens a ctx.ui.select menu with cached status, environment-variable presence, and registered model counts (IDs stay hidden). Detail views show one short display segment per page (long records continue without losing text); Previous/Next navigate between segments and Back returns to the sections. Close and cancellation leave without running cursor-agent, checking auth, discovering models, or writing configuration. Environment values, paths, account details, and credentials are never shown. Credentials stay in the external CLI/environment, and model-file configuration stays external; this menu has no editable provider settings. Outside the TUI, bare /cursor-pi keeps the existing status-command behavior.

Installation verification

On every session_start the extension verifies the local Cursor CLI before you can use it:

1. Runs cursor-agent --version. If the binary is missing or unusable, Pi shows a warning with install guidance (curl https://cursor.com/install | sh) instead of failing later mid-conversation.
2. Runs cursor-agent status. If you are not logged in, Pi warns you to run cursor-agent login.

The same checks run on demand via /cursor-pi verify and /cursor-pi status. Stream failures during a turn also include this setup guidance.

Configuration

| Environment variable | Description | | -------------------- | ----------- | | CURSOR_PI_BIN | Override the Cursor CLI executable path. Defaults to cursor-agent. | | CURSOR_PI_MODELS | Comma- or space-separated model ids to register. Defaults to auto,composer-2.5,gpt-5.3-codex-high,claude-sonnet-5-thinking-xhigh,gemini-3.7-flash-high. | | CURSOR_PI_TIMEOUT_MS | Per-turn cursor-agent -p timeout in milliseconds. Defaults to 300000. | | CURSOR_PI_CONTEXT_WINDOW | Override the advertised context window in tokens. Defaults to 272000. |

Example:

CURSOR_PI_MODELS="auto,gpt-5.3-codex-xhigh-fast" pi

How it works

For each Pi model turn, the extension:

1. Serializes Pi's system prompt, conversation transcript, and available tool schemas into one text prompt.
2. Spawns the local Cursor CLI with cursor-agent -p --output-format text --model <selected> --mode ask --trust.
3. Writes the serialized prompt to Cursor over stdin.
4. Converts Cursor stdout into a Pi assistant text message, or converts <pi_tool_call>{...}</pi_tool_call> markers into native Pi tool calls.
5. Emits a clear assistant error if the CLI is missing, exits non-zero, is aborted, or times out.

Cursor runs in read-only ask mode so its own tools never edit files or run shell commands. Pi tool schemas are included in the prompt, and explicit <pi_tool_call>{...}</pi_tool_call> markers are handed back to Pi so Pi executes tools through its normal pipeline.

Notes and limitations

• This is slower than native HTTP providers because a cursor-agent -p process starts for each model turn.
• Tool calling is prompt-bridged with <pi_tool_call>{...}</pi_tool_call> markers, so it is less reliable than native provider tool calling but still keeps execution in Pi.
• Text input only; image content is omitted from the serialized prompt.
• Usage/cost numbers are estimated from character counts because the text transport does not report token usage.
• The default model list is static; new Cursor model ids appear after setting CURSOR_PI_MODELS (see /cursor-pi models for available ids).