MCP servers that do one thing well
The best MCP servers I have worked with are not clever. They expose narrow tools, return boring data, and leave enough trace evidence that a human can understand what happened.
Read articleFaiz Writes
Six months of engineering notes about MCP servers, OpenClaw, AI security, agent workflows, and the parts of shipping software that only become clear after the build gets real.
Latest
MCP servers that do one thing well
Tool boundaries, traceable actions, and why boring interfaces make better agents.
OpenClaw
Repeatable local builds April notes
Security checklist
Prompt boundaries
Tool permissions
Audit trails
OpenClaw notes from a repeatable build
What started as a local setup became a checklist for making agentic workflows easier to reproduce.
Read articleAI security starts with boring boundaries
Most AI safety work in small tools is not abstract policy. It is permission design, logging, and refusing to blur trust zones.
Read articleSix-month archive
Mock posts seeded to give the blog a lived-in engineering rhythm.
May 10, 2026
MCP servers that do one thing well
MCP servers
April 18, 2026
OpenClaw notes from a repeatable build
OpenClaw
March 27, 2026
AI security starts with boring boundaries
AI security
February 21, 2026
A practical audit trail for AI tools
Agents
January 16, 2026
Shipping static previews on Cloudflare Pages
Deployment
December 8, 2025
Local agents, remote services, and where I draw the line
Workflow