MCP Experience Differences: Claude vs ChatGPT
denis_gvozd · reddit · 2026-07-14
The author tested the performance of the same OAuth-protected remote MCP service in Claude and ChatGPT, finding that protocol-level availability is consistent, but the client experience differs significantly.
Commonalities
- Same remote MCP endpoint
- Same OAuth authentication and access control
- Same read-only toolset: searchknowledge, fetchdocument, getentitycontext, listavailablesources
- Both target knowledge retrieval scenarios
Main Differences
- In Claude, MCP feels more like a natural capability extension within the chat experience
- In ChatGPT Pro, private custom MCP requires Developer Mode, making the product experience feel more like a "testing interface"
- Tool execution, logs, and request details are more exposed, which isn't natural for regular users
Author's View
MCP clients should distinguish between two lifecycles:
- Development Phase: Register endpoints, complete OAuth, inspect tools, refresh schema, debug requests, view logs
- Usage Phase: Approve MCP, use directly in normal chat/Projects, keeping technical details folded away
The author believes private MCPs shouldn't have to become publicly listed apps just to reach a "normally usable" state. They also mention they are building a private knowledge layer for a real estate team; this isn't a pure demo, but a UX problem in actual deployment.
More from coding & agent
- GitHub review bot hits its PR limit and forces a 39-minute cooldown — DanielLockyer · 2026-07-22
- Max reasoning effort appears to be mobile-only in Codex Remote, not desktop — GabGarrett · 2026-07-22
- A Reddit demo argues online stores should expose carts and pricing through MCP — gelembjuk · 2026-07-22
- Open-source AI SDK provider routes Vercel apps through a local Codex subscription — lgrammel · 2026-07-22
- Codex vs Claude Code: Which Is More Popular? — jxnlco · 2026-07-22
- CodeRabbit uses layers, diagrams and a chat agent to rethink code review — _jaydeepkarale · 2026-07-22