Giving every agent every MCP tool backfired, so they built virtual servers
Background-Job-862 · reddit · 2026-07-26
A team says exposing every tool from every MCP server to every agent turned out to be a mistake.
They had about 15 MCP servers registered, and each connected agent could see the full tool list. That caused two problems:
- Tool selection got worse as the list grew to 80+ tools; the model started choosing vaguely similar tools from the wrong server.
- Security scoping was too broad because agents could see tools they had no business touching.
Their fix was to introduce a virtual MCP server:
- Curate a small, scoped subset of tools from multiple real servers.
- Present that subset as one server per team, workflow, or agent.
- Example: a customer-support bot sees only ticketing and CRM tools, not billing or infra tools.
Unexpected upside: tool selection improved, and security reviews became much easier because the virtual server’s tool list clearly defines what the agent can do.
More from coding & agent
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- Investment Analyst Asks How to Build a Claude-Based Diligence Agent Stack — Careless_Tie2286 · 2026-09-11