Connecting to multiple MCP servers in one client

1. Find tool function meta datas
2. Construct tools list for OpenAI chat API, e.g.

{
    "type": "function",
    "function": {
        "name": "add_numbers",
        "description": "Adds two numbers.",
        "parameters": {
            "type": "object",
            "properties": {
                "a": {"type": "integer", "description": "First number"},
                "b": {"type": "integer", "description": "Second number"},
            },
            "required": ["a", "b"],
        },
    },
},

3. Call OpenAI chat/response API, with tools param, and tool_choice optionally
4. Get response content and tool calls if it exists

msg = response.choices[0].message
resp, tool_calls = msg.content, msg.tool_calls

5. Extract tool function name and tool function arguments from the tool calls

function_name = tool_call.function.name
function_args = tool_call.function.arguments
if function_args:
    function_args = json.loads(function_args)

6. Find MCP server session by function name. SSE server URL may require sse path: http://127.0.0.1:8081/sse

async def call_sse_server(self, function_name, arguments=None):
    async with sse_client("http://127.0.0.1:8081/sse") as (sse_read, sse_write):
        async with ClientSession(sse_read, sse_write) as sse_session:
            # Initialize the SSE server connection
            await sse_session.initialize()

            # Call tools and resources on the SSE server
            reverse_result = await sse_session.call_tool(
                function_name, arguments=arguments
            )
            return reverse_result

Initialize server session, cache it, and close it gracefully on exit

7. Extract tool function execution result from server session.call_tool.
   res = esult.content[0].text

Refactor MCPAgent, connect to MCP servers on class initialization, keep sessions in attribute sessions, keep function name to session mapping, close session on exit