ai agent

README.md

@@ -196,6 +196,22 @@ reports truncation instead of silently dropping results.
 Discovery is per project (`~/.ra3modxml/endpoints/<project>.json`), so several
 VS Code windows can enable agent access at the same time without shadowing each
 other, and a client can never be silently answered from a different project.
+Each window also registers itself under `~/.ra3modxml/instances/`, and a small
+merged `~/.ra3modxml/index.json` lists the live instances and their project
+roots. A window that crashes is cleaned up by whichever instance starts next —
+no workspace has to be reopened first.
+
+**No Node installation is required.** The launcher runs the bundled server on
+VS Code's own Electron binary (`ELECTRON_RUN_AS_NODE=1`) and only falls back to
+`node` from `PATH` if that binary is missing. The launcher is rewritten on every
+activation, so updating or moving VS Code does not break an existing MCP config.
+
+The MCP server is plain JSON-RPC over stdio, so an agent can also query the
+index without any MCP setup at all by piping a request into the launcher. A
+matching CLI (`cli.js`, next to `mcpServer.js` in `dist/agent/`) answers from
+the live index when VS Code is running and from the exported snapshot
+otherwise; commands that need element context say so explicitly rather than
+returning an empty result.
 
 See `docs/ai-agent-integration-plan.md` for the full design and progress.