Sentry asks: all tier-1 harnesses now support MCP tool search — drop our own primitives?
zeeg · x · 2026-11-04
Sentry's David Cramer poses an architecture question: with every tier-1 agent harness now supporting MCP tool search and deferred tools, should they drop Sentry's internal tool search primitives and expose every tool directly?
A getsentry/toolkit doc (PR #1862) maps harness support:
- Tier 1 (must work): Codex, Claude Code, Cursor — desktop and CLI/TUI counted as separate surfaces
- Tier 2 (should not break): OpenCode, Pi, VS Code + Copilot Chat
- Codex has deferred every MCP tool behind Responses API toolsearch since 0.143.0 (needs GPT-5.4+); Claude Code's ToolSearch is on by default and always defers; Cursor IDE has used dynamic context discovery since 2.4, CLI lazily loads since 2026-09
Signal: deferred tool loading / tool search has shifted from optional optimization to a native tier-1 harness capability, forcing middleware layers to re-evaluate their own abstractions.
Related event: Sentry CEO Questions Need for Custom Tool Search in the MCP Era(3 posts)→
More from coding & agent
- GhostDraw: a hobbyist turns hand-drawn sketches into app diagrams with tldraw and AI — Dhananjay_Tech · 2026-11-04
- CoworkOS Debuts: Three-Layer Agent Coworker Framework Lifts Qwen3.6 by 9.76 Points — KyeGomezB · 2026-11-04
- DOLORES: Training-Free Meta-Scaffold Lifts 8B Models Past 32B Baselines — StellaLisy · 2026-11-04
- Berkeley-affiliated StarTrail open-sources a deep research benchmark and paper-search toolkit — CShorten30 · 2026-11-04
- AI job-search framework built on Claude Code hits 69.3k GitHub stars — adnan_hashmi · 2026-11-04
- DuplexAgent-RSI separates live speech from async agent work in voice assistants — jiqizhixin · 2026-11-04