Use agent memory with ChatGPT and OpenAI Agents

Durable memory that outlives the chat: store facts, search them, recall them later. Connect ChatGPT and OpenAI Agents to the WingMan MCP gateway once and these tools are available in every conversation.

1. Connect ChatGPT and OpenAI Agents

With the OpenAI Agents SDK (Python), connect over Streamable HTTP as below. The same URL works as a remote MCP tool in the Responses API and as a custom connector in ChatGPT developer mode.

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


async def main():
    async with MCPServerStreamableHttp(name="wingman", params={"url": "https://wingmanprotocol.com/mcp"}) as wingman:
        agent = Agent(name="Assistant", mcp_servers=[wingman])
        result = await Runner.run(agent, "Use the wingman tools for this task.")
        print(result.final_output)


asyncio.run(main())

# Responses API instead of the SDK:
# tools=[{"type": "mcp", "server_label": "wingman", "server_url": "https://wingmanprotocol.com/mcp", "require_approval": "never"}]

2. The tools you get

Names and descriptions come straight from the live tool registry.

• store_memoryPersist a value across your instances: PUT /memory/{ns}/{key}.
• recall_memoriesSearch both recall notes AND memory entries for content related to your query.
• search_memoryFull-text search over YOUR memory values using FTS5.
• summarize_memoryCondense ALL entries in a namespace into a single markdown summary via local Llama 3.2 3B (free, no token cost).
• share_memoryShare a memory namespace with another handle.

These tools are in the default tool list, so they show up as soon as the server connects.

3. Ask for it

Give ChatGPT and OpenAI Agents a task that needs it, for example “use store_memory from wingman”. The tool call goes to https://wingmanprotocol.com/mcp and the result comes back into the conversation.

Agent memory with other clients

More tool categories for ChatGPT and OpenAI Agents