/Nexus/
A production MCP server connecting Claude AI to a client's entire internal stack — CRM, database, calendar, and Slack — through natural language.
Oliver's operations team was toggling between five tools to complete basic workflows. We built a production MCP server that exposed custom tools to Claude — letting the AI query sales data, book meetings, update HubSpot, post Slack notifications, and generate pipeline reports, all through a single chat interface. Manual dashboards became optional overnight.
/Project Goals/
Connect every critical business tool to one AI interface so the team can run daily operations through conversation, not clicks.
The client's team was losing two hours daily switching between HubSpot, Google Calendar, Slack, and internal databases. The MCP server needed to expose tools that were fast, reliable, and secure — with JWT authentication, rate limiting, and Redis caching ensuring performance at scale. The goal was 80% of routine operations handled via AI chat within 30 days.
/Results/
80% of daily operations now run through AI chat. 15 active users. Zero downtime since deployment across six months.
5 custom MCP tools went live serving 15 daily active users. Average tool response time held under three seconds. The team eliminated four manual reporting workflows entirely. The client described the shift as moving from operating software to simply talking to their business. Uptime has held at 99.9% since launch with no incidents logged.
MCP ProtocolClaude APIPythonFastAPIPostgreSQLRedisDockerAWS