Role intent · distinct from geo Kenya page
Claude MCP developer for production agents
I build Claude-powered agents that use scoped tools against real operational systems — with evals, verification, and cost control — not chat demos. Based in Nairobi (UTC+3), remote-friendly for EMEA and async US teams.
Managed agent sessions that survive interruptions — the pattern behind WaybillAgent warehouse walks.
MCP-style tool boundaries, idempotent writes, and the same RLS discipline as multi-tenant ops data.
Outcome-driven evaluation and AIDC transcript controls before customer-facing changes ship.
Proof on this site
- WaybillAgent — Claude Opus vision agent for warehouse audits
- Outcome-driven evals — Hive fork for business-result scorecards
- Writing: warehouse copilots, LLM safety for AIDC
- Geo context: AI engineer in Kenya (different intent — location proof, not MCP tooling)
FAQ
Production systems where Claude uses scoped tools (MCP or equivalent) against real systems of record — WMS, ERP, tickets — with logging, evals, and human gates for risky writes.
The work is systems engineering: tool schemas, auth boundaries, long-horizon sessions, cost controls, and outcome evals — not one-off prompt paste.
WaybillAgent (Claude Managed Agents + vision for warehouse audits), the Hive outcome-eval fork, and writing on copilots and AIDC safety on this site.
Contact: hi@josephrwanda.com