grove

IDE bridge: /ide for external agents

Run claude in your own terminal, type /ide, and pick Grove — exactly the way you'd connect Claude Code to VS Code or Cursor. Grove implements the same editor-integration protocol those IDEs do, so an unmodified claude CLI gets the integration it expects, with Grove as the host editor.

What you get

  • Selection sync. The CLI can see what you have selected in Grove — your cursor position in source files, and even text selections in rendered markdown notes, mapped back to source lines. Ask claude about "this function" with the function selected in Grove, and it knows what you mean.
  • Send to chat. The selection toolbar's "send to chat" button mentions the selected file#lines into the connected CLI's prompt. This is a deliberate gesture — passive cursor movement is shared as context but never types into your prompt.
  • Editor control. The CLI can open files in Grove and jump to lines — "show me where this is defined" lands in your editor.

How discovery works

On launch Grove stands up a local WebSocket server and advertises it in the discovery directory the claude CLI scans (~/.claude/ide/), with a per-session auth token readable only by your user. The CLI's /ide picker matches its working directory against Grove's open workspace, so it offers Grove when you're in the right repo. Switch workspaces in Grove and the advertisement follows.

Bridge vs. MCP

These are two different integrations that compose:

  • The MCP server gives agents Grove's tools — comments, notes, tasks, code navigation.
  • The IDE bridge makes Grove a host editor for an agent running outside Grove — selection context and file-open control.

An external claude connected via /ide can use both at once.

Current limitations

  • Diagnostics aren't forwarded yet (the CLI sees an empty list).
  • Open-editor and dirty-file queries return empty — selection and file-open are the live surfaces.
  • One workspace folder at a time; no connection-status indicator in the UI yet.