Multi-agent personal platform
Four specialist agents — fitness, nutrition, grocery and finance — running on NestJS and exposed to AI assistants through a single MCP server.
- Role
- Architect & sole developer
- Type
- Personal project
- Timeline
- [YYYY] — [STATUS]
- Stack
- NestJS · TypeScript · MCP
How the pieces connect
The problem
[What wasn’t working before — the scattered apps, spreadsheets or habits this platform replaces, and why a single assistant interface mattered.]
The approach
Each domain gets its own agent with its own tools and data, built as a NestJS module. Rather than wiring every agent into every assistant, the platform exposes them all through one MCP server, so any MCP-capable client can use them. [Add how agents share context or hand off to each other.]
What I learned
[Two or three honest lessons — what you would design differently, what surprised you about agent boundaries or tool design.]
Key decisions
One MCP server for all agents
A single interface keeps client setup to one connection and lets agents be added without touching the assistant side.
Trade-off: [What this made harder]One agent per domain
Fitness, nutrition, grocery and finance stay independent, so each can change its tools and prompts on its own.
Trade-off: [Cross-agent coordination cost]NestJS as the host
[Why NestJS — modules, dependency injection, familiarity, existing tooling.]
Trade-off: [What you gave up]