GPT Lookup — phase-2f-c-mcp-production-patch-design — 2026-05-14

Query

phase-2f-c-mcp-production-patch-design-2026-05-13

Located source

• knowledge/current-state/reports/phase-2f-c-mcp-production-patch-design-2026-05-13.md
• Title: Phase 2F-C — MCP Production Patch Design Report (2026-05-13)
• Status in source: Investigation + design only. No code changes, no rebuild, no restart, no commit.

Key findings

1. list_documents latency is Python-side full-table materialization; recommended patch 2F-D1 is SQL pushdown plus partial text_pattern_ops index.
2. MCP transport blocker is app-level content negotiation, not nginx; recommended patch 2F-D2a is JSON/SSE response negotiation.
3. move_document is intentionally deprecated/501 but still advertised; recommended patch 2F-D3 removes it from MCP schema/dispatcher while preserving HTTP 501 endpoint.
4. Write latency is dominated by inline vector sync duplicated by the existing PG-NOTIFY listener; recommended patch 2F-D4 makes vector indexing async/eventual-consistent.
5. Public GPT endpoint policy should remain /gpt-mcp/{token}/mcp -> /mcp-gpt with 8 tools; do not expose /mcp-gpt-full publicly.

Recommendation preserved from source

Implement 2F-D1 + 2F-D2a before reconnecting ChatGPT, then 2F-D3, and land 2F-D4 separately after production verification.