Dynamic tool discovery vs loading every MCP schema upfront for agents
Pakiya09 · reddit · 2026-11-16
A developer highlights how agents waste context: hooking up 4-5 MCP servers dumps thousands of tokens of tool descriptions and JSON schemas into every turn, mostly for tools never used — hurting small-context models and long multi-step runs.
His fix: runtime tool discovery, where the agent searches a catalog and only pulls the full schema when it's about to call that tool. The monid project's CLI and MCP integration is built this way.
He asks the community whether on-demand schema loading actually saves enough tokens and improves tool selection, or whether the extra search step adds annoying latency.
More from coding & agent
- Quant researcher co-writes a short math-heavy book on agentic engineering with LLMs — blaizedsouza · 2026-11-16
- llm-council built into Hermes: multiple models answer, blind-rank, and a chairman decides — Scobleizer · 2026-11-16
- Grok Build v1.0.87 improves multi-agent workflows, MCP panel and session search — XFreeze · 2026-11-16
- Swarm orchestration runs 12 parallel agents to build a Blender 3D model, claims tester — venturetwins · 2026-11-16
- AgentWeb lands YC W27 spot building endpoints that let AI agents operate any site — damnGruz · 2026-11-16
- Prototype pairs kids' code, a real robot and Snap Spectacles AR for learning — stspanho · 2026-11-16