AgentRAM + Codex

Codex supports MCP servers through its config file, so adding AgentRAM's memory is a few lines of TOML. Your agent then stores and recalls named memory across sessions.

What you get

Named, persistent memory across Codex sessions, shared memory for multi-agent workflows, and no vector database to run. Memory becomes tools your agent calls directly. The concepts are covered in the MCP integration overview.

Prerequisites

• An AgentRAM API key. Create one free.
• Codex CLI installed, and Node.js 18+ (node -v).

Add the server

Codex reads MCP servers from its config file (~/.codex/config.toml). Add an entry:

[mcp_servers.agentram]
command = "npx"
args = ["-y", "agentram-mcp"]
env = { AGENTRAM_API_KEY = "agentram_your_key_here" }

Config location and format follow Codex's MCP support. If it differs in your version, check the Codex docs. The command and env var are the same.

Quick start

Start a Codex session and try it:

Store the staging URL as "https://staging.example.com" under the key staging_url.

Later, in a new session:

Recall staging_url from AgentRAM.

Smoke test and troubleshooting

• Tools do not appear. Restart Codex after editing config.toml, and confirm the TOML is valid.
• 401 or auth error. Check the AGENTRAM_API_KEY value in the config starts with agentram_.
• npx cannot find agentram-mcp. Node is older than 18. Check node -v.