Building MCP servers
Several projects where agents maintain their own data — through custom MCP servers with clear tools, permissions and mandatory sources.
Ten things I actually do with AI — not experiments, but in production, with guardrails you can explain to data protection, audit and management.
Several projects where agents maintain their own data — through custom MCP servers with clear tools, permissions and mandatory sources.
One agent enters, another verifies. Nothing goes live without evidence and a source.
Parallel sessions with agreements, handovers and clear ownership instead of chaos in the repo.
Secrets only as references, cost approval before paid calls, human sign-off before anything public.
Claude Code and Codex day to day: reviews, tests, CI — so the team gets faster, not less sure.
RAG and knowledge bases that give agents and people the same answers.
PDFs, minutes and emails turned into clean data — with review where it matters.
Local models and transcription when data must not leave the building.
Keywords, landing pages and ads prepared by agents, approved by a human.
Evals and cost control: what a model can do and what it costs is measured, not estimated.
30 minutes, free. Afterwards you know whether and how I can help.