MCP Tool Schemas Are Eating Your Context
Ok-Tooth1667 · reddit · 2026-07-12
The author noticed that connecting multiple MCP servers to an agent causes tool selection accuracy to steadily decline as tool count increases, showing a near-linear correlation with the number of servers.
The root cause: every tool's name, description, and full JSON Schema are serialized into the context during each request. For example, 4 servers with 60 tools consume about 11–12k tokens just for definitions, while the system prompt itself is under 2k.
Their practical fix wasn't reducing the total number of tools, but reducing the number of tools visible to the model at any given time:
- Exposing a lightweight search/load step first.
- Letting the agent determine which tools are needed.
- Loading only 3–5 relevant tools into the context.
Results:
- Tool block size dropped from 11–12k tokens to 2k.
- Routing accuracy improved.
- Latency decreased.
- Overall context pressure was significantly relieved.
The author concludes that the real bottleneck isn't the total number of tools, but the volume of serialized schemas in the active context.
Related event: MCP Tool Schema Overload Impairs Agent Accuracy(3 posts)→
More from coding & agent
- Hugging Face wraps 6-month Training Agents series with open recordings and code — SergioPaniego · 2026-10-26
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26