ads-mcp v0.4.0

ads-mcp v0.4.0 提供 49 个 MCP 工具，较 v0.3.0 新增 17 个，覆盖三条主线：

• 原理图 → 原生 Layout → Momentum → emModel → 电路联合仿真：从已有原理图生成原生版图，配置网格、扫频与原生模型输出，再将 emModel 接回原理图进行联合仿真。
• 参数化 EM 联合优化：原理图 VAR 驱动 Subnetwork 形式参数与参数化 Layout；未缓存尺寸触发 Momentum 重算，由原生 Optim 评估目标。
• vendor 模型接入：提供 AVX 0402、3.9 pF 电容的官方下载脚本、厂商库注册、型号选择与 RF 仿真验证说明，不分发厂商库。

新增工具涵盖设计参数、原生 Layout 与符号、厂商库、emModel 与缓存、Momentum 网格与求解配置、PySide6 窗口管理、网格显示、优化器释放及 S 参数比较。Layout Lookalike Symbol 使用 ADS 原生 AEL 函数 momSymbol_createLayoutLookAlikeSymbol 生成，同时支持 Dual/Quad 矩形符号。已有仿真工具扩展 GUI 联合仿真与 Optim，并完善完整 N 端口 Touchstone 导出。

附带两个独立 ADS 示例工程：联合仿真反射式移相器、微带贴片天线 EM 优化；另附 vendor 模型下载与接入示例。106 项单元与回归检查通过，完成原生 emModel 联合仿真、参数化天线优化及未缓存尺寸重算验证。

移相器是对 IEEE Access 2019 论文 180°单比特电路的大致复现，1.4 GHz 两态相差约 179.94°。PIN 使用校准的非理想模型，L/C 仍为理想元件，结果属于仿真验证。

实测环境：Windows、ADS 2027 v6.5.0、ADS 自带 Python 3.14，需有效 ADS/Momentum 许可证；其他版本和平台尚未验证。

详见 中文 README、English README 和 新增工具说明。

ads-mcp v0.2.0 — Topology library + Schematic audit

New tools (23 → 25)

• ads_audit_schematic — KiCad eeschema-style schematic overlap audit: severity-graded thresholds per collision class (body / text / wire), connectivity-aware body-body exemption, ERC-style dangling wire-endpoint detection
• ads_autolabel — annotation auto-placement: candidate sides (right/left/up/down, least displacement first), moves text only, never components

LC topology library — demo/lc_synth.py + demo/build_lc_filters.py: Butterworth/Chebyshev g-parameter synthesis → 8 verified topology cells (LPF/HPF/BPF/BSF × 2 responses). Chebyshev LPF in-band ripple 0.10 dB, Butterworth cutoff exactly 1.00 GHz, zero schematic collisions per cell.

Docs — tool surface/highlights/roadmap/verification synced to 25 tools; ADS version statement changed to a verified-only claim (tested on 2027 v6.5.0, API since 2025U2); KiCad eeschema acknowledgment added.

ads-mcp v0.1.1 — Chebyshev synthesis + ADS optimizer closed loop

Design-loop upgrade: proper filter synthesis plus ADS's own optimizer, driven headlessly.

New tools (21 → 23)

• ads_add_goal — optimization goals (expression, min/max bound, frequency band, weight)
• ads_add_optim — optimization controller (gradient/hpVMO, random, quasinewton, …)
• ads_add_var — optional opt_min/opt_max turns any variable into an optimization variable

New demos

• demo/chebyshev_synth.py — g-parameter synthesis (N=5, 0.1 dB, fc=1 GHz) → stepped-impedance microstrip on FR4
• demo/run_synth.py — synthesized design verified: −3 dB cutoff 1.030 GHz (target 1 GHz), passband worst −0.80 dB
• demo/run_optimize.py — full optimizer closed loop: gradient (hpVMO), 26 iterations / 336 evals, error function 0.149 → 0.0072

Optimized results (FR4, 5-element SI-LPF)

• cutoff 1.050 GHz
• passband S21 worst −0.61 dB (was −0.80)
• in-band S11 worst −15.8 dB (was −11.9)
• stopband S21 @1.5 GHz −18.0 dB (was −4.3)

Engine fixes: repeated parameters take list values; goal frequency bands rendered as plain Hz; EnableOptim=yes emitted when an optimization controller is present (hpeesofsim silently skips optimization otherwise); docs updated to 23 tools.

ads-mcp v0.1.0

Initial public release.

21 MCP tools for the headless Keysight ADS design loop:
schematic construction (guardrailed) → netlist → hpeesofsim (async task id + polling) → S-parameter results, Touchstone export, PNG plots, GUI hand-off.

Verified on ADS 2027 (v6.5.0): conversational build of a 5-element step-impedance microstrip LPF on FR4 — passband −0.08 dB, −3 dB cutoff 1.025 GHz (target 1 GHz), −38 dB rejection @ 6 GHz.

See README for setup (requires ADS 2025U2+ with bundled Python 3.14).