6 MCP servers eat a huge chunk of context before you even type — lessons from a tool-bloat experiment
sandeepK1125 · reddit · 2026-11-19
A Reddit user measured how 6 connected MCP servers (GitHub, filesystem, Postgres, Slack, browser, docs search) consume a large share of the context window with tool names, descriptions, and JSON schemas alone.
Key observations:
- With 60+ tools, the model increasingly picks close-but-wrong tools
- Verbose documentation-style tool descriptions add up fast
- Overlapping servers (two that both "search") cause the most mistakes
Mitigations that helped:
- Separate server profiles per task (coding vs research) instead of one big config
- Disabling unused tools where the client allows it
- Preferring servers with a few broad tools over dozens of narrow ones
The author also asks the community at what tool count quality degrades and whether short tool descriptions should be the server's or client's job.
Related event: Too Many MCP Servers Bloat Context; Devs Share Slimming Tips(2 posts)→
More from coding & agent
- Google's GUARD paper: frozen base model, self-editing agent harness gains +8.89 points — omarsar0 · 2026-11-19
- Microsoft open-sources openclaw-dev: one-command hosted OpenClaw agent on Azure with sandboxed tool execution — adnan_hashmi · 2026-11-19
- 6-minute Hubble documentary made with Claude and Jimeng CLI workflows — xiaohu · 2026-11-19
- Coding agents gone wild: test-cheating, scope drift, and rogue DB migrations — Common_Dream9420 · 2026-11-19
- Regenerative software: keeping intent as the source of truth so code can always be regenerated — gamagamer · 2026-11-19
- Every on Claude Fable 5.2: Human Thinking at the Start Matters More as AI Builds More — every · 2026-11-19