Question: Accessing Request Context in MCP Tool Handlers

[antoncxx]

I’m building an MCP server and have an authentication middleware that validates a session token from the request headers.
Here’s the relevant middleware code:

pub async fn authentication_middleware(
    State(context): State<AppContext>,
    headers: HeaderMap,
    mut request: Request<axum::body::Body>,
    next: Next,
) -> Result<Response, StatusCode> {
    let session_token = extract_token(&headers).ok_or(StatusCode::UNAUTHORIZED)?;

    let db_token = get_db_token(&context)
        .await
        .ok_or(StatusCode::INTERNAL_SERVER_ERROR)?;

    let session = fetch_session_details(&context, db_token, &session_token)
        .await
        .ok_or(StatusCode::INTERNAL_SERVER_ERROR)?;

    if matches!(session.r#type, RemoteAccessType::Mcp) {
        request.extensions_mut().insert(session);
        Ok(next.run(request).await)
    } else {
        Err(StatusCode::BAD_REQUEST)
    }
}

The middleware works fine, but I need to access the session (inserted into the request extensions) inside my tool handler.
However, in all examples I’ve seen, tool handlers only receive their parameter schema, like this:

#[tool(description = "Execute a command on the client machine")]
async fn execute_command(
    &self,
    Parameters(ExecuteCommandParameters { command, arguments }): Parameters<ExecuteCommandParameters>,
) -> Result<CallToolResult, ErrorData> {
    // I need to access the session here
}

How can I access the session data (inserted in request extensions by middleware) from within a tool handler?
Is there a supported way to read per-request context or extensions from within a tool?