What 2,800+ Claude Code issues reveal about AI dev tools teams actually use
We analyzed 2,800+ Claude Code issues. Here are four themes that separate demos from durable AI dev tools—plus concrete wins teams can ship now.
AI dev tools live or die on operational details: how they resume on day 14, how they behave under rate limits, and how safely they cross version upgrades. After analyzing 2,800+ issues in anthropics/claude-code, four themes surfaced over and over—the same seams that either stall adoption or turn a clever demo into a reliable teammate.
Here’s the punchline:
- Standardize and integrate: Adopt ACP/AGENTS.md and ship first‑class IDE support (Zed/Positron) so agents behave the same across tools.
- Safer MCP by design: Per‑agent tool filtering, isolation, and elicitation reduce noise and improve reliability.
- Durable operations: SDK + M2M auth, quota‑aware fallback, named sessions, watch and schedule make long‑running tasks boringly reliable.
- Clarity in the UI: Inline usage and consistent theming/input affordances keep developers in flow.
Methodology: We filtered for open feature requests in anthropics/claude-code and ranked demand using a popularity score: score = (2 * 👍) + (0.5 * comments) + (1 * participants, approximated as 1 + min(comments, 5)). We then grouped requests into four themes and selected representative issues for each.
Quick overview
- Integrations, IDE support, and standardization
- MCP configuration
- Interoperability and long-run tasks
- UI/UX and visibility improvements
Integrations, IDE support, and standardization
Problem: Claude Code lacks a shared protocol and agent config standard across editors, leading to bespoke integrations and inconsistent behavior.
User quote
“Without supporting this standard, Claude Code may miss the opportunity to be easily integrated into a growing ecosystem of ACP-compatible editors.” — #6686 — ACP support
Top asks
- Add ACP agent/server compatibility for editor integrations — #6686
- Support AGENTS.md alongside CLAUDE.md — #6235
- Ship first-class Zed integration (via ACP) — #1471
- Recognize Positron as a supported VS Code fork — #1731
- Clarify stance on “own editor” vs integrations — #282
MCP configuration
Problem: Per-agent control over MCP tools (filtering, isolation, elicitation, notifications) is missing, creating noisy or unsafe defaults and brittle workflows.
User quote
“Elicitation allows MCP servers to request additional information from users during interactions…” — #2799 — MCP Elicitation
Top asks
- Per-agent allow/deny filtering of MCP tools — #7328
- Agent-scoped isolation for tools and contexts — #4476
- Implement MCP elicitation flow — #2799
- Notify when available MCP tools change — #4118
- Disable/hide specific MCP functions — #6759
Interoperability and long-run tasks
Problem: Teams need SDK/M2M auth for services, quota-aware fallback, named/resumable sessions, and automation (watch-files, scheduled runs) for durable operations.
User quote
“Currently, Claude Code sessions are only identifiable by auto-generated session IDs and timestamps.” — #2112 — Custom session naming
Top asks
- Ship SDK + M2M auth — #6259, #1454
- Implement quota-aware fallback — #2944
- Support named sessions — #2112
- Add watch-files — #418
- Add scheduled actions — #2794
UI/UX and visibility improvements
Problem: Users need inline usage/limits visibility and consistent theming/ergonomics (Shift+Enter behavior, pre-send editing) to stay in flow and build trust.
User quote
“Let the user see … where they are in the context usage at all times.” — #516 — Always show available context percentage
Top asks
- Inline usage display and per-project/org metrics — #516, #1109
- Auto light/dark theme parity — #2990
- Terminal/editor theme parity presets — #1302
- Pre-send edit for pasted/generated text — #3412
- Shift+Enter newline ergonomics — #1758
How agyn helps
We focus on the operational layer—approvals, observability, and continuity—so teams can run AI coding workflows with fewer surprises. If you’re building or adopting AI dev tools, let’s compare notes; we can share patterns, dashboards, and guardrails that have scaled.