persistent mcp servers

[merimus]

Question

Trying to use a mcp server, in particular one which provides memory.

servers = load_mcp_servers('mcp_config.json')
agent = Agent(model, toolsets=servers)

agent.run() appears to create an instance of the mcp each time I make a call so it doesn't retain memory.
Is there a way to cause it to keep the mcp server around.

In this case it's configured as a docker container.

Additional Context

I am using the sequantialthinking mcp server
First call is to create a series of thoughts, works fine and create a series of thoughts.
Second call is to query those thoughts but there doesn't appear to be any content.

[DouweM]

@merimus You can start the MCP servers around the context where you use them repeatedly using async with agent:.

[merimus]

This is roughly what I am doing.


{                                                                                                       
  "mcpServers": {                                                                                       
    "sequentialthinking": {                                                                             
      "command": "docker",                                                                              
      "args": [                                                                                         
        "run",                                                                                          
        "--rm",                                                                                         
        "-i",                                                                                           
        "mcp/sequentialthinking"                                                                        
      ]                                                                                                 
    }                                                                                                   
  }                                                                                                     
}     

servers = load_mcp_servers('mcp_config.json')                                                           
model = OpenAIChatModel(                                                       
    'openai/gpt-oss-20b',                                                                               
    provider=OpenAIProvider(                                                                            
        base_url='http://10.0.0.16:1234/v1', api_key='your-api-key'                                     
    ),                                                                                                  
)     

    async with Agent(model,                                                                             
                     model_settings=settings,                                                           
                     toolsets=servers) as planningAgent:                                                
        result = await planningAgent.run(                                                               
            "Use sequential thinking to create a step by step plan for an undergraduate compilers cours\
e.",                                                                                                    
            usage=usage, usage_limits=limits)                                                           
        print(result.output)                                                                            
                                                                                                        
        result = await planningAgent.run(                                                               
            "Show all thoughts in sequantialthinking",                                                  
            usage=usage, usage_limits=limits)                                                           
        print(result.output)

[merimus]

Watching docker... the instance running the mcp server does NOT get restarted.