I build practical AI-agent infrastructure, automation systems, and security-focused developer tools.
Current focus: MCP security, local-first agent workflows, and tools that make autonomous software agents safer to use with real systems.
Local-first firewall for MCP tool calls.
- Blocks poisoned MCP tool descriptions before agents can see them
- Detects descriptor drift after the first trusted baseline
- Rate-limits runaway tool calls
- Scans tool inputs for shell injection, path traversal, and XSS-like payloads
- Writes redacted audit logs for tool-call visibility
Local memory backbone for multi-agent systems using SQLite, knowledge graphs, and fast retrieval.
Screen-aware MCP experiment for giving agents local desktop context and capture-based memory.
- Security and observability for AI-agent tool use
- Local-first agent infrastructure that does not depend on broad cloud access
- Developer tools that turn messy automation workflows into reliable systems
Most of my work is built in public through small, shippable projects: CLIs, MCP servers, workflow engines, dashboards, and automation systems. I care about demos that run, tests that prove behavior, and tools that solve operational problems.