A 5-layer setup gives Claude Code, Codex and Cursor the same shared memory

Arindam_1729 · x · 2026-10-21

Developer Arindam shares a practical 5-layer setup for shared context across Claude Code, Codex, Cursor and OpenCode.

The problem: every agent session starts from zero, and AGENTS.md only holds rules, not knowledge — runbooks, API docs, decisions and their rationale live elsewhere, so agents reading the same AGENTS.md still give different answers.

The five layers (with a decision table in the images):

Knowledge layer: OpenLore runs as an MCP server (localhost:8080) exposing an in-memory shell (cat/grep/find/jq) over a docs/ tree (how-to, architecture, api, incidents). The article includes exact MCP setup commands for Claude Code, Codex, Cursor and OpenCode, plus SSH access (port 2222) for CI jobs to grep the same docs.

Related event: Five-Layer Architecture Gives Claude Code, Codex and Cursor Shared Agent Memory(2 posts)→

Original post →

More from coding & agent

coding & agent channel →