Does Codex CLI have memory?

Yes — native Memories (local files under ~/.codex/memories, off by default, /memories to toggle) plus AGENTS.md for repo instructions. Native Memories do not sync to Cursor or Claude, and they were unavailable in the UK/EEA/Switzerland at launch.

Why are Codex memories unavailable in the UK?

OpenAI gated native Memories out of the EEA, UK, and Switzerland at launch. Teams there use AGENTS.md, skills, and third-party MCP memory. ReadmeDB is one of those MCP options and is not tied to Codex's regional memory flag.

AGENTS.md vs Codex memories?

AGENTS.md is the committed briefing Codex (and other tools) read. Native Memories extract facts from past Codex chats into a local store. Neither is a shared, human-readable project brain other editors can write.

How do I share Codex context with Cursor or Claude?

Point Codex, Cursor, and Claude Code at the same ReadmeDB MCP namespace. Agents write decisions.md and open-tasks.md; the next tool reads them. Do not expect ~/.codex/memories to appear in Cursor.

readmedbmcp

Solution · Codex CLI

Codex CLI memory that Cursor and Claude can actually read

Codex can remember inside Codex. Native Memories live under ~/.codex, start off, and were blocked in the UK/EEA at launch. AGENTS.md is a static briefing. ReadmeDB is the markdown Codex writes and your other agents open tomorrow.

What Codex already gives you

AGENTS.md — native. /init scaffolds it. Cross-tool if you keep it as rules, not a diary.
Native Memories — extract from prior chats into local files. Toggle with /memories. One machine, one Codex home. Not a team wiki.
Session resume — continue a transcript by id. Useful for the same run, not for “what did we decide last month.”

If you are in the UK or EU, assume native Memories may simply not be there. Plan the shared layer anyway.

Connect Codex in 30 seconds

Add to ~/.codex/config.toml (or project .codex/config.toml on a trusted repo). Streamable HTTP uses url, not command:

[mcp_servers.readmedb]
url = "https://app.readmedb.com/api/mcp"
http_headers = { "Authorization" = "Bearer rdb_your_key" }

Prefer an env var? Use bearer_token_env_var = "READMEDB_API_KEY" instead of a static header. Then /mcp in the TUI. Ask Codex to create decisions.md and open-tasks.md.

Native memories vs the project brain

	Codex Memories	ReadmeDB
Where	~/.codex/memories	Cloud namespace, versioned files
Who reads it	Codex on that machine	Codex, Cursor, Claude, Windsurf, CLI
UK / EEA	Often unavailable	Works — it is just MCP + markdown
You can open it	Local extracts, not a PR-able doc	Plain .md in the dashboard

Use both if Memories are on

Native Memories for Codex-only leftovers. ReadmeDB for anything another tool or person must see. If Codex is extracting MCP tool output back into Memories, turn that off so you do not duplicate the brain.

Start free — Codex + Cursor on one brain →

AGENTS.md Cursor Claude Windsurf Project brain CLI