Lucas Gerads demonstrates MCP servers for oscilloscope and SPICE simulator integration with Claude Code
Lucas Gerads has developed a workflow that integrates Claude Code with hardware development tools through MCP (Model Context Protocol) servers. The setup connects Claude Code to both a LeCroy oscilloscope and a SPICE simulator, creating a feedback loop between simulation and physical hardware verification.
Workflow and Approach
Gerads experimented with using Claude Code for hardware development, initially trying to describe circuits in natural language prompts. He found this approach worked for trivial circuits but became difficult for complex designs. The breakthrough came when he gave Claude Code access to his oscilloscope and SPICE simulator, allowing immediate feedback during development.
Specific Implementation Details
The system includes three main components:
lecroy-mcp: MCP server for LeCroy oscilloscopesspicelib-mcp: MCP server wrapping spicelibrc-filter-demo-files: Demo setup files from the accompanying video
Practical Lessons Learned
Gerads shares several specific implementation guidelines from his experience:
- Oscilloscope integration: Claude doesn't see your physical setup, so don't let it guess what's connected where. Ensure Claude never gets stale measurement data. Don't dump raw data into Claude's context - save it to a file and let Claude interact with it indirectly.
- Microcontroller workflow: Give Claude a pinout/pinmux map explicitly. Prepare a Makefile that exposes functions like build, flash, ping, and erase, and encourage Claude to rely on it. Claude should not construct these commands on the fly.
Use Cases Demonstrated
The setup has proven valuable for:
- Validating SPICE circuits and models
- Embedded programming tasks
- Data analysis automation (particularly time axis normalization and data alignment, which were previously tedious manual processes)
While the demo uses a trivial circuit and MCU to illustrate the approach, Gerads notes that the workflow scales well to more complex circuits and real embedded projects.
📖 Read the full source: HN AI Agents
👀 See Also
OpenClaw Context Meter Plugin Shows Telegram Token Usage Percentage
A new OpenClaw plugin displays token usage percentage after every Telegram bot response, showing values like '45k / 200k (22%)' and detecting compaction events. The plugin avoids OOM issues by hardcoding context windows instead of using execSync.
HN SOTA: Tracking Coding Model Popularity via Hacker News Comments
HN SOTA is a daily updated pipeline that scrapes Hacker News comments, uses Gemini to identify coding models from the OpenRouter list, and logs sentiment per mention into a public Google Sheet. The top 10 models by mentions over a 10-day trailing window are displayed.
Open Source System Captures Claude Code Patterns into Evolving Documentation
Developer Lee Fuhr has released three open source repositories that systematically capture and codify learnings from working with Claude Code. The system includes a methodology document with 14 principles and 19 patterns, an architecture classification framework, and a memory system with 149 features.
Ninetails Memory Engine V4.5: Int8 Quantization + LRU Cache Cuts Local MCP Memory to 60MB
The Ninetails Memory Engine V4.5 uses Int8 scalar quantization and LRU cache eviction to reduce vector storage from 6KB to 1.5KB per embedding, keeping the entire engine at 40-60MB RAM. It combines 70% vector similarity with 30% BM25 search in a fully local SQLite implementation.