Move the language server into devtools, wire the VS Code extension, and show anonymous closures (eve_update = function) by frame prototype instead of a shifted CallGraph zip.
Co-authored-by: Cursor <cursoragent@cursor.com>
Adds 'eve mcp': a desktop-only headless MCP host (stdio by default, --port for TCP, --root for the project). It runs the full runtime, attaches DevTools (eve_eval / eve_run_script / snapshot), and starts the new EditorHost module that renders JSON-defined editor Views (ImGui) bound two-way to Squirrel ViewModel tables (MVVM), persists editors under editors/, and exposes the eve_host_* tool family (apply/state/set_value/events/capture/script/shutdown...). The stdio transport uses a background stdin reader and keeps stdout JSON-RPC-only; script print() is routed to stderr.
Tests: stdio handshake + tools/list round trip, host editor binding (VM<->View sync, set_value -> VM write + onChange events, save/reload persistence). Docs updated (AI与MCP支持.md, 命令行设计.md, tools/eve-mcp README) with the editor JSON protocol and Codex stdio setup.
Add the scene_director Squirrel kit (src/scripts/scene_director.nut) with
spawn/move/scale/rotate/remove/material/lighting/camera/cameras/info/modify/
reset, embedded into the engine and exposed to host games as
eve.sceneDirectorScript. New MCP tools (eve_scene_director_install/status,
eve_scene_reset, eve_scene_modify, eve_scene_info, eve_camera_generate) are
thin dispatchers over the kit, so agents can place props, set lighting,
frame cameras, inspect truth and reset the stage through MCP alone.
The kit's cross-references are written explicitly as scene_director.<field>
to avoid Squirrel table-closure his resolution, and cameras() uses a
fixed 6-way compass table (no math-lib dependency) so it works both in the
engine and in plain ssq test VMs.
Adds a vision-model bridge (devtools/RenderVision) that captures the
current frame (Graphics readback -> PNG data URL) and sends it with
engine render parameters to an OpenAI-compatible vision model over
plain HTTP, returning prose that describes the frame and explains the
render-parameter <-> visual-result correspondence for other LLM agents.
- New MCP tools:
- eve_render_describe (fresh/reason; cached otherwise)
- eve_render_vision_config (set/read config, key masked)
- Breakpoint hit and runtime-error pause auto-flag a pending dump that
McpServer::poll performs on the main (render) thread.
- Config via env (EVE_VISION_BASE_URL/API_KEY/MODEL/PATH/TIMEOUT_MS)
or the config tool.
- No TLS/OpenSSL is linked, so the endpoint must be plain HTTP (works
with local Ollama/vLLM/LM Studio).
- Compiles cleanly on MSVC (C++20); tests extended for new tools.
Add 18 new MCP tools to the embedded McpServer so agents can drive
the live game beyond the debugger surface:
- scene/entity query + transform mutation
- procgen recipe enumeration, map grid generation, mesh build
- 2D physics world create/list/raycast/remove
- render status + frame screenshot (PNG)
- particle emitter spawn, audio master control
Tool handlers use ModuleManager to reach Scene/Procgen/Physics/
Graphics/Particles/Audio and degrade gracefully when a module is
not mounted. Extend the tools/list and unit-test coverage.
Raw `new Object()` values stored in Dynamic::Var broke Stringifier with
BadCast during tools/list. Wrap nested objects/arrays in Ptr, emit
compact JSON for newline framing, and convert request fields safely.
Co-authored-by: 西风逍遥游 <sunxfancy@gmail.com>
Expose a localhost MCP JSON-RPC server (--mcp-port) alongside DAP so AI
agents can pause, eval, snapshot, and slice errors on a live session.
Add DevTools AI session log/ImGui panel (eve.dev.ai, F9) and a Cursor
stdio bridge under tools/eve-mcp.
Co-authored-by: 西风逍遥游 <sunxfancy@gmail.com>
西风逍遥游