Codex Central Memory Setup

Summary#

Each Codex host should use Closed Akashic MCP directly as its shared memory. Do not clone or depend on local agent-knowledge for normal work.

Host Setup#

Configure Closed Akashic authentication at the machine level, outside any project repository.
Add the closed-akashic MCP server to ~/.codex/config.toml.
Put the host-level memory rules in ~/.codex/AGENTS.md.
Start a new Codex session so the MCP registration and AGENTS instructions are loaded.
Smoke-test with search_notes, read_note, and a small safe write-back.

Required MCP Endpoint#

[mcp_servers.closed-akashic]
url = "https://knowledge.openakashic.com/mcp/"

Preferred Auth for Shared Local Agent Setups#

Prefer a static Authorization header in the machine-local Codex config when the host is shared by local agents and GUI/session environment propagation is unreliable:

[mcp_servers.closed-akashic]
url = "https://knowledge.openakashic.com/mcp/"
http_headers = { Authorization = "Bearer <redacted>" }

Do not store or document the real token value in repositories or project notes.

Env Var Fallback#

Use bearer_token_env_var only when the token must not be stored in ~/.codex/config.toml:

[mcp_servers.closed-akashic]
url = "https://knowledge.openakashic.com/mcp/"
bearer_token_env_var = "CLOSED_AKASHIC_TOKEN"

For shell-based hosts, keep the token in a host-local profile or service environment:

export CLOSED_AKASHIC_TOKEN="set-your-master-token-here"

Already-running GUI apps or agent processes can keep stale env state, so restart the Codex session after changing auth config.

Required AGENTS File#

Use Codex AGENTS Template as the canonical text for ~/.codex/AGENTS.md.

For a single deployable instruction file, use Codex MCP Deployment.

For token-specific setup details, use Closed Akashic Shared Agent Token Setup.

Copy-Paste Setup#

Run this on each Codex host after configuring authentication outside any project repository.

mkdir -p ~/.codex
touch ~/.codex/config.toml

grep -q '^\[mcp_servers.closed-akashic\]' ~/.codex/config.toml 2>/dev/null || cat >> ~/.codex/config.toml <<'TOML'

[mcp_servers.closed-akashic]
url = "https://knowledge.openakashic.com/mcp/"
# Preferred for shared local setups:
# http_headers = { Authorization = "Bearer <redacted>" }
# Fallback when using a host environment variable:
# bearer_token_env_var = "CLOSED_AKASHIC_TOKEN"
TOML

Then write ~/.codex/AGENTS.md from Codex AGENTS Template.

Operating Flow#

Search Closed Akashic before substantial work.
For validated public knowledge, use search_akashic when relevant.
Read the matching project index README.
Read relevant project repo docs.
Work in the repo or server.
Write back one concise note or append one focused section.

Project Folders#

Use bootstrap_project with optional folders when a project needs a custom shape.

Example folder sets: - product app: architecture, playbooks, incidents, decisions, reference - research project: papers, experiments, datasets, prompts, reference - ops service: runbooks, deployments, incidents, dashboards, reference

Reuse#

This replaces agent-knowledge. Project-specific AGENTS.md files may still exist, but they should add local rules only, not create another memory system.