AgentRAM + the OpenAI Agents SDK

The OpenAI Agents SDK has a first-class MCP integration, so you can launch AgentRAM's memory server as a subprocess and hand its tools to any agent. A few lines of Python and your agents remember across runs.

What you get

Persistent, named memory your Agents-SDK agents can read and write across runs, shared memory for multi-agent teams, and no vector database to manage. Memory is exposed as tools the agent calls. See the MCP integration overview for the underlying model.

Prerequisites

• An AgentRAM API key. Create one free.
• Python 3.10+ and the SDK: pip install openai-agents (a virtual environment is recommended).
• Node.js 18+, since the AgentRAM server runs through npx.

Connect the server

The Agents SDK can launch an MCP server as a subprocess with MCPServerStdio. Point it at agentram-mcp:

import asyncio
from agents import Agent, Runner
from agents.mcp import MCPServerStdio

async def main():
    async with MCPServerStdio(
        name="AgentRAM",
        params={
            "command": "npx",
            "args": ["-y", "agentram-mcp"],
            "env": {"AGENTRAM_API_KEY": "agentram_your_key_here"},
        },
    ) as agentram:
        agent = Agent(
            name="Assistant",
            instructions="Use AgentRAM to store and recall facts across sessions.",
            mcp_servers=[agentram],
        )
        result = await Runner.run(agent, "Remember that my name is Sean, then confirm.")
        print(result.final_output)

asyncio.run(main())

The agent now has AgentRAM's store and recall tools available and will use them as the task requires.

Smoke test

Run the script once to store a value, then run a second run that recalls it in a fresh process to prove the memory persists across runs, not just within one conversation.

Troubleshooting

• ImportError for agents. The virtual environment is not active, or the SDK installed into a different interpreter. Confirm which python points into your venv.
• 401 or auth error. The AGENTRAM_API_KEY in env is missing or wrong; it should start with agentram_.
• npx cannot find agentram-mcp. Node is older than 18. Check node -v.